Evaluation of asphalt mixtures modified with low-density polyethylene and high-density polyethylene using experimental results and machine learning models | Scientific Reports

Scientific Reports volume 14, Article number: 24601 (2024) Cite this article

Metrics details

The widespread use of low-density polyethylene (LDPE) and high-density polyethylene (HDPE) plastics has resulted in a large amount of waste plastic that requires appropriate disposal or reuse. One potential solution is to use them in the modification of asphalt concrete (AC) mixtures for more sustainable highways. To study this possibility, permanent deformation and dynamic modulus (DM) of the LDPE and HDPE modified AC mixtures was investigated by conducting flow number (FN), flow time (FT) and DM tests on Superpave gyratory compacted specimens. Machine learning models; multi-layer perceptron (MLP), radial basis function neural network (RBFNN), generalized regression neural network (GRNN) and support vector machine (SVM) were used to predict the DM on the basis of frequency and temperature parameters. The model’s performance was gauged by analyzing the root mean square error, mean relative error, and coefficient of determination. The study findings revealed that the LDPE and HDPE modified AC mixtures provide 2.07 times and 1.27 times better resistance to permanent deformation, respectively, than their counterpart. It was also found that the LDPE and HDPE modified AC mixtures have 2.1 times and 1.4 times higher DM values, respectively, than the Control AC mixtures. Among the machine learning models, MLP (R2 = 0.98) showed best accuracy in predicting DM and thus is recommended to be used in similar studies due to its robustness. Additionally, the feature importance analysis revealed that frequency has the highest impact on DM predictions, followed by temperature and the inclusion of the LDPE.

Plastic materials have attracted worldwide attention because of their numerous applications, low recycling rates, and environmental concerns associated with their disposal methods. In 2015, the world’s annual plastic production was 407 million tons, which is nearly two-thirds of the world’s population1. Unfortunately, only 45% of the manufactured plastic is recycled or burned, with the remaining 55% being dumped in landfills2. The global climate and its inhabitants are under severe threat from the improper management of landfilled plastic. To lessen these landfill and environmental concerns, there is a critical need to investigate additional alternative uses for leftover plastic.

Pakistan has produced 3.9 million tons of waste plastic in 2020. Approximately 70% of this waste plastic is mismanaged, either left in landfills or dumped on land and water bodies across the country. This makes Pakistan one of the leading countries with poor waste plastic management services3,4. In August 2019, the plastic-bag initiative free program was launched to ban the plastic bag’s production, sale and purchase in the country5. Numerous other approaches have also been investigated to increase the recycling rate of waste plastic. The construction sector, in this regard, offers a large opportunity to properly manage a huge amount of waste plastic by utilizing them in asphaltic pavements. This approach is effective not only in reducing the waste plastic but also in improving the performance and serviceability of asphalt pavements, which are the most widely known pavement types globally6.

There are different forms of plastic, but low-density polyethylene (LDPE) and high-density polyethylene (HDPE) have acquired prominence due to their frequent use in daily life. Common applications of LDPE and HDPE include but are not limited to plastic bags, containers, trays, agricultural film, bottles, toys, and pipes. According to a previous investigation7, LDPE has a melting point of 108 °C, a density of 0.914 g/cm3, a breaking strength of 7.5 MPa, and a breaking elongation of 400%, while HDPE has a melting point of 131 °C, a density of 0.954 g/cm3, a breaking strength of 21 MPa, and a breaking elongation of 500%. In most studies, these materials were added to the asphalt mixes by dry or wet process7,8. In the dry process, the plastic is initially introduced into the heated aggregate, mixed properly, and then a binder is added to produce the required asphalt mix. In the wet process, a modified binder, prepared by adding plastic to the binder, is introduced into the heated aggregate to prepare the required asphalt mix.

The permanent deformation in asphalt concrete (AC) mixtures is of great interest to researchers aiming to mitigate rutting in asphaltic pavements9. Stiff binders/mixtures, specially modified with polymers, plays an important role in limiting the permanent deformation of AC mixtures. According to a previous study10, 5% LDPE-modified asphalt mixtures provide the highest resistance to permanent deformation. While another study11 revealed that 6% LDPE-modified stone mastic asphalt (SMA) mixtures provide 96% higher resistance to permanent deformation compared to the base SMA mixtures. Following the similar trend, HDPE-modified asphalt mixtures have also been shown to be more resistant to permanent deformation at high temperatures12. In a comparative study, it was concluded that the AC mixtures modified with HDPE polymers outperform base AC mixtures in terms of permanent deformation13. However, asphalt mixtures with higher percentages of LDPE and HDPE were found more vulnerable to permanent deformation14.

The dynamic modulus (DM) of AC mixtures is widely employed in the mechanistic-empirical design and structural analysis of asphaltic pavements15. The findings of a research study revealed that AC mixtures modified with 6% LDPE offer greater resistance to repeated tire loading16. Another study17 set a restriction of 4% for the maximum usage of plastic that can improve the resilience modulus of asphalt mixtures. Likewise, the SMA mixtures with 10% polyethylene terephthalate (PET) exhibit high modulus values at high frequencies and low temperatures18. Similarly, laboratory testing of high-density polyethylene-dense-graded asphalt mixtures revealed higher values of resilient modulus13. In addition, due to compatibility concerns, adding HDPE plastic to asphalt mixtures beyond the limit of 5% of the aggregate weight causes a decrease in the fatigue and resilient modulus values19.

In a comparative study20, it was found that the asphalt mixtures treated with plastomeric polymers, styrene-butadiene-styrene (SBS), and rigid recycled plastic containing graphene were found to have equivalent permanent deformation, fatigue, and stiffness. Another study concluded that at low loading frequencies, asphalt mixtures modified with recycled plastic outperform the crumb rubber-modified asphalt mixtures in terms of DM, flow number (FN), and rutting resistance21. Likewise, a recent study found that asphalt mixtures modified with optimal LDPE provide more resistance to moisture damage and permanent deformation than the asphalt mixtures modified with optimal heavy automobile waste tyres22.

Traditionally, Witczak’s predictive equations and other statistical techniques have been used to predict the DM of AC mixtures, but more recently, modeling with machine learning has become a viable alternative due to the issues of the accuracy and explicit mathematical forms of the predictive equations23,24. The performance of AC mixtures has been predicted in previous studies using a variety of machine learning methods, with neural networks being the most comprehensive one. The neural network performs significantly better than the empirical equations in terms of predicting accuracy25,26,27,28,29,30. The neural network could accurately predict the DM of AC mixtures with a remarkable value of the coefficient of determination31,32,33. Following the same pattern, it was discovered that neural networks outperformed regression-based models in accurately representing the effects of the binders’ rheological properties, the mixture’s volumetric characteristics, and reclaimed asphalt pavement on the DM of AC mixtures34. Even though neural networks can give accurate estimates of the DM of AC mixtures, it takes a lot of work to construct a useful network, even for a small data set.

The artificial neural networks (ANNs) are further sub-classified into different types, based upon the processing of the neurons in the hidden layers. Three types of ANNs are employed in the present study, namely, multilayer perceptron (MLP), radial basis function neural network (RBFNN), and generalized regression neural network (GRNN). The difference between them is the processing of weights in the hidden neurons. More detailed information about these types can be found in35,36.

Along with the neural networks, another study37 has shown that the support vector machine (SVM), which is considered as a computationally expensive classical technique under the machine learning domain, is a useful tool for estimating the DM of AC mixtures. A recent study38 conducted on predicting the viscoelastic characteristics of AC mixtures revealed better predicting performance of SVM than the relevance of vector machine and random forest algorithms. The use of ANNs and SVM models to predict the performance of AC mixtures has revealed that some studies39,40 considered ANNs to be a better prediction technique, whereas others41,42,43 considered SVM to be a better prediction technique. This study has thus aimed to use both techniques for the prediction of the DM of AC mixtures.

The alteration of AC mixtures with waste plastic is not a new field; nonetheless, the application of cutting-edge analysis approaches could not only improve the insights and technical reliability of this technology but could also aid in the prediction of AC mixture performance. In this study, an attempt has been made to thoroughly analyze the permanent deformation and DM of the LDPE and HDPE modified AC mixtures by executing three nominated tests (FN, FT and DM) of national cooperative highway research program (NCHRP), which is rarely practiced with its complete protocol for the mentioned materials. The study has placed more emphasis on evaluating the effectiveness of a broad set of machine learning models in predicting the DM of the LDPE and HDPE modified AC mixes. Four machine learning models, namely MLP, RBFNN, GRNN, and SVM have been trained to predict the DM of the Control and modified AC mixtures. At the end, the prediction accuracy of these models was tested using several performance indicators such as root mean square error, mean relative error, and coefficient of determination.

The materials and testing methods used in this study are briefly discussed in the following subsections.

Attack refinery limited (ARL), Pakistan’s pioneer in crude oil refining, provided 60/70 penetration grade binder for this research study. The properties of the chosen binder were confirmed by a number of laboratory experiments, which are detailed in Table 1.

This study made use of limestone aggregate, while stone dust was utilized as filler. Several laboratory tests were performed to determine the basic properties of aggregate, such as elongation index and flakiness index (ASTM D4791), LA abrasion (AASHTO T96), soundness (AASHTO T104), and sand equivalency (AASHTO T176). The aggregate was found to have an elongation index of 7%, a flakiness index of 5%, LA abrasion of 26%, soundness of 3.38%, and a sand equivalent value of 72%. All these findings are within the specified limits of the test standards. In addition, the aggregate blends used in this study were developed following the gradation curve shown in Table 2.

Waste plastic bags (LDPE) and plastic bottles of soft canes (HDPE) were selected to be used for the modification of AC mixtures. The LDPE was manually shredded into smaller sizes, while the HDPE was shredded into sizes equivalent to or smaller than 1.18 mm in a jaw crusher. An overview of the properties of the LDPE and HDPE are provided in Table 3.

Superpave mix design method was used to fabricate AC specimens in replicates with binder contents of 3.5, 4, 4.5, 5.0, and 5.5%. Following the 4% air voids criteria44, the optimum binder content (OBC) for the Control AC mixtures was determined in a way to satisfy the strength and volumetric requirements. For determining the optimum polymer contents (OPCs), a dry method of mixing was used for incorporating the shredded LDPE and HDPE into AC mixtures. In the dry method of mixing, the aggregates were first heated to a temperature of 180 °C. The designated quantity of the shredded waste plastic was then uniformly distributed over the coarser portion of aggregates. The mixture was then conditioned in a hot air oven at 180 °C for a duration of 2 h. Following the conditioning period, the partially melted shredded waste plastic was blended with the aggregates for a period of 1–2 min to achieve partial coating. Once adequate mixing and coating had been confirmed visually, bitumen heated to the requisite mixing temperature of 160 °C was introduced and mixed thoroughly with the plastic-coated aggregates17. The waste plastic was incorporated into aggregates at varying percentages of 4, 6, 8, 10, 12 and 14% by the weight of the OBC to ensure proper coating. Gyratory compacted specimens of the modified AC mixtures were then prepared, and based on the maximum stability criterion, the optimum proportions of the HDPE and LDPE were determined to be 10% and 12%, respectively. Results of the Superpave mix design for the Control and modified AC mixtures are summarized in Table 4. Using the optimum proportions of binder content and polymer contents, triplicate specimens for performance testing were prepared and tested as explained in the following sections.

In this study, FN, FT and DM tests were conducted on Superpave gyratory compacted specimens. The FN and FT tests were conducted in compliance with the AASHTO TP 79 standard. Both the tests were conducted at a stress level of 300 kPa and an effective temperature of 54 °C. The FN test has a stress period of 0.1 s and a recovery period of 0.9 s, and it ends after 10,000 cycles or when a specimen underwent a cumulative strain of 5%.

The DM test was conducted on Superpave gyratory compacted specimens in replicates in accordance with the AASHTO TP 62. In the DM test, specimens were subjected to a continuous compressive sinusoidal loading over a temperature range (4.4–54.4 °C) and frequency sweep (0.1–25 Hz). The test data (Appendix E) was utilized for assessing and prediction of the DM of asphalt mixtures.

The FN test was conducted both on the Control and modified AC specimens to evaluate their resistance to permanent deformation. In this test, specimens were subjected to repeated loads, and the resulting axial strains were measured. This test most accurately simulates actual field loading conditions due to the rest period that is allowed in-between the load applications. The FN test determines the resistance of an asphalt mixture to permanent deformation by measuring cumulative axial strains as a function of the loading cycles. The test’s outcome is a cumulative strain versus loading cycle curve, which typically have primary, secondary, and tertiary flow zones. Depending on the properties of the asphalt mixtures, each zone begins with a different cycle. In the primary zone, the strain rate decreases as more loading cycles are completed, whereas in the secondary zone, the strain rate remains constant and in the tertiary zone, the strain rate increases as more loading cycles are completed.

The FN test was conducted at 54.4 °C because permanent deformation is more critical at elevated temperatures. The test was set to finish either after completion of 10,000 cycles or when a specimen underwent a cumulative strain of 5%, whichever came first. Figure 1, represents accumulated strain versus load cycle curves, from which it can be inferred that modified AC mixtures require higher number of cycles to achieve the three stages mentioned above, and thus can withstand more load applications to permanent deformation. More specifically, the LDPE-modified AC mixtures performed better as it yielded 2.07 times and 1.63 times higher FN values than the Control and HDPE modified AC mixtures, respectively. This suggests that the LDPE-modified AC mixtures are more resistant to permanent deformation, which is in line with the findings of various studies10,11,21,22. This behavior of the LDPE-modified AC mixtures could be attributed to their high stiffness values, which make them more rigid and resistant to permanent deformation.

Effect of loading cycles on accumulated strains under flow number test.

The permanent deformation of AC mixtures was also estimated by plotting the compliance and time on a log scale and determining the regression coefficients ‘a’ and ‘b’, known as the compliance parameters45. In general, higher values of compliance parameters represent a higher susceptibility of an asphalt mixture to permanent deformation46. In this study, the secondary portion of the resulting curves were used to estimate the compliance parameters because these parameters do not account for material deformation in the tertiary zone. In comparison to the Control AC mixtures, the LDPE and HDPE modified AC mixtures were found to have low compliance parameter values, indicating higher resistance to permanent deformation.

Asphalt mixture performance tester (AMPT), was used to perform the FT test at a stress level of 300 kPa and test temperature of 54.4 °C. The accumulated strain was compared and analyzed after 10,000 cycles. According to the findings presented in Fig. 2, all the mixtures have exceeded the maximum cycle limit of 10,000 cycles. Thus, the comparison is made on the basis of accumulated strains. It can also be inferred that the LDPE-modified AC mixtures accrue fewer strains and are less prone to permanent deformation than the Control and HDPE modified AC mixtures. However, compared to the FN test, there is less of a difference in the accumulated strain for the Control and modified AC mixtures46.

Effect of loading cycles on accumulated strains under flow time test.

The FT test is also associated with the workability of AC mixtures. A decrease in FT corresponds to increased workability of AC mixtures, leading to fewer challenges during the pavement construction phase. Thus, Fig. 2 also demonstrates that adding the LDPE and HDPE to AC mixtures improves their workability, potentially resulting in optimal compaction and better pavement quality. The better performance of the LDPE AC mixtures compared to the HDPE AC mixtures in terms of permanent deformation can be accredited to the ability of the LDPE coating the aggregate particles effectively and resulting in sufficient film thickness. However, the long-term performance of both types of the waste plastic modified asphalt mixtures becomes almost equal47,48.

AMPT was used to perform the DM test at six frequencies (25, 10 5, 1, 0.5, and 0.1 Hz) and four temperatures (4.4, 21.1, 37.8, and 54.4 °C). This test was carried out with a compressive sinusoidal load to simulate actual traffic conditions. According to the test findings shown in Fig. 3, a rise in temperature from 21.1 to 37.8 °C, caused 61% decrease in DM values, on average, for the LDPE, HDPE and Control AC mixtures, respectively. The reason could be the softening of asphalt binder at higher temperatures, which also reduces the stiffness of the asphalt mixtures and causes the DM to decrease. It was also found that the sweep of frequency from 25 to 0.1 Hz, caused 78% decrease in DM values, on average, for the LDPE, HDPE and Control AC mixtures, respectively. This could be due to the fact that low frequency cyclic loading involves larger strain amplitudes and as the strain amplitude increases, the DM of the asphalt mixtures decreases.

Dynamic modulus of AC mixtures.

Furthermore, the DM values for the LDPE-modified AC mixtures were found to be 2.1 and 1.5 times greater than the Control and HDPE modified AC mixtures, respectively. The DM values for the HDPE-modified AC mixtures were also found to be 1.4 times higher than that of the Control AC mixtures. As a result, both the LDPE and HDPE AC mixtures are found to be stronger than their counterparts16,17,18,21.

The experimental data had three variables observed, i.e. temperature, frequency and dynamic modulus. In addition, the type of sample (Control, HDPE-modified or LDPE-modified) was represented with three dichotomous variables (0, 1) included with each sample. Finally, the machine learning models were developed for predicting DM using 5 independent variables. These models provide a convenient way to model the properties of asphalt mixtures with modifications for which the existing analytical methods do not remain valid. As mentioned above in Sect. 1, these techniques have been used to good effect, in previous studies for similar applications. One of the prominent features of machine learning algorithms is to learn from the data without any prior knowledge about the relationships between parameters49. There exists a vast array of machine learning models, but this study utilizes ANNs and SVM for their proven successful history for modeling prediction problems50,51 and the previous successful examples of their use in similar problems related to AC mixtures.

Highway pavements are an important and expensive part of infrastructure. Hence, increasing their durability and reducing their cost is a crucial aspect to attain sustainable growth. This study investigates the performance of polyethylene modified AC mixtures on the basis of rarely used three candidate test (FN, FT and DM) results. The findings of this study would encourage the design of long-lasting pavements with recycled material, thus, creating a positive impact on the economy and environmental aspects of highway infrastructure pavement. Moreover, machine learning algorithms were developed for predicting the observed performance parameters. The following conclusions were drawn.

The results of the FN test, which measures permanent deformation, revealed that the LDPE-modified AC mixtures exhibit 2.07 times and 1.63 times higher values, on average, of FN compared to the Control and HDPE AC mixtures, respectively. Furthermore, the comparison of the HDPE-modified and the Control mixtures showed that the average FN value of the HDPE mixtures is 1.27 times higher than that of the counterpart.

The FT test also supports these results by indicating that the LDPE-modified AC mixtures accrue fewer strains and are less prone to permanent deformation compared to the Control and HDPE modified AC mixtures. However, the difference in accumulated strain between the Control and modified AC mixtures is lower than that observed for the FN test.

The FT test also suggests that adding the LDPE and HDPE to AC mixtures improves their workability, potentially resulting in optimal compaction and better pavement quality.

The DM values of the LDPE-modified AC mixtures were found to be 2.1 times and 1.5 times higher, on average, compared to the Control and HDPE modified AC mixtures respectively. In addition, the HDPE-modified AC mixtures were found to have DM values 1.4 times higher, on average, than that of the Control AC mixtures.

Among the machine learning techniques used in this study, MLP showed best performance by demonstrating error less than 20% and R2 more than 0.9 for both training and test datasets. Hence, due to these results, the use of these models is highly recommended, and it is expected that other researchers can use these models, with the specifications provided in this study. However, future research may focus on employing ensembles of machine learning models which have proven useful for other cases.

The feature importance analysis shows that frequency has the highest impact on the DM prediction, followed by temperature and inclusion of the LDPE. Some of these observations are also confirmed by previous studies. Inclusion of the LDPE results in higher DM, hence, it is a preferable modification compared to the HDPE. However, it was found that all types of AC mixtures showed very low DM in extreme temperatures (low or high).

Future research studies might concentrate on investigating the long-term endurance of the LDPE-modified AC mixtures at extreme temperatures. Moreover, studying the effect of extreme temperatures on the microstructural and rheological properties of the LDPE and other waste plastic modified binders and mixtures may offer more insights about their performance mechanisms under thermal stresses. Findings from such research studies may direct the development of more lasting and sustainable pavements.

Four types of machine learning models were used in this study, namely MLP, RBFNN, GRNN and SVM. MLP is the most used type of ANN. They are made up of three types of layers wherein signals are transferred forward after processing from the previous layer. The power of MLP is the multiple parallel processing of inputs using non-linear functions which could be of any type possible52.

RBFNN is a type of ANN that is found to be more efficient computationally, compared to MLP. Its training process can be divided into two parts, unsupervised and supervised. The unsupervised part consists of determination of center and range (widths) of Gaussian functions for the hidden neuron, while the supervised part consists of linear weight determination for the neurons53.

GRNN is another type of ANN which works on a one-step training process resulting in faster development of nonlinear input-output models. In addition to that, another advantage is the ability to model multi-input-multi-output relationships which makes it better than traditional regression models. However, their performance is highly impacted by the outliers and irrelevant variables in the dataset. Hence, it is recommended to be used with a filtering tool like principal component analysis54. Lastly, SVM is a technique that works on classifying the available sample space into an appropriate number of hyperplanes. Hence, the algorithm is simple, flexible and, most importantly, has the ability to provide insights into the impacts of certain samples on the predictions55.

For all types of ANNs, the hyperparameters were chosen after a trial-and-error procedure. In that case, the available dataset was divided into three parts using random sampling. One part of the data was used for training the data which consisted of 50% samples, 25% was used for validation in the process of tuning the hyperparameters while the rest was kept as a test dataset. For the SVM model, the dataset was simply divided into training (75%) and test (25%) since no hyperparameters were optimized in that case. In this procedure, the model accuracy was observed for different combinations of these parameters and, in the end, the model with the highest accuracy on the test dataset was selected. Furthermore, all variables were normalized using a minimax function to avoid the problem of local minima, which has been associated with ANNs as evident from past studies56.

The MLP consisted of 3 layers with 4 neurons in the hidden layer. Activation function for the input layer was linear, while it was hyperbolic and logistic for the hidden and output layer, respectively. The accuracy parameters for this model are shown in Table 5 while Fig. 4, shows the plot of predicted and observed values.

Plot of observed and predicted values for MLP.

The RBFNN model had 3 layers with 7 neurons in the hidden layer. The activation function for the input and output layers was linear, while that for the hidden layer was exponential as shown in Fig. 5.

Plot of observed and predicted values for RBFNN.

The GRNN model consisted of 4 layers, with 2 hidden layers, having 36 neurons in the first hidden layer and 2 in the second. The activation function for all the layers was linear, except for the first hidden layer which had exponential function as shown in Fig. 6.

Plot of observed and predicted values for GRNN.

The SVM divided the input space into 24 hyperplanes, called support vectors. As per the data, it was a regression model, with 5 independent variables. The kernel function was radial basis, and all hyperplanes were kept unbound due to the variation in datasets. Accuracy parameters for SVM (as shown in Table 5) indicate a close similarity between the training and test dataset values, while Fig. 7, shows the plot of predicted and observed values.

Plot of observed and predicted values for SVM.

In this research, SVM and different types of ANN approaches were used to predict the DM response of the HDPE and LDPE modified AC mixtures, based on two input variables. Different parameters were used to test the accuracy of the prediction models for the training and test datasets. A comparison of these parameters is shown in Table 5. These parameters showed that both the techniques (ANNs and SVM) demonstrate satisfactory ability in capturing the variability of the output, as shown by the coefficient of determination value. It is reiterated at this point that the same samples were used for training and testing each model, respectively, which were selected randomly for each case. Hence, the comparison of accuracy is expected to be valid and robust for the available dataset and the case in study. More specifically, MLP was the best in predicting the outcome for the test and training datasets, followed by the SVM, as shown in Table 5. Another important trend is the difference in accuracies for training and test datasets for any model. A larger difference shows overfitting of the model, while a closeness between the accuracy on training and test datasets shows robustness of the model. The hyperparameters for each of these models were optimized based on the accuracy of the test dataset. Their values are shown in the previous sub-sections, which could be beneficial for researchers to replicate the results of this study.

Accuracy statistics for MLP models are observed to be very high and similar for the training and test dataset. This could be an indication about the prediction power and robustness of the model for predicting dynamic modulus for the available sample of this study. Several other previous studies have shown similar trends for accuracy of MLP57,58, which justifies the application of MLP models over traditional models for normal asphalt mixtures. The accuracy parameters shown in Table 5 also indicate that the SVM model had a satisfactory accuracy level, but it was lower than the MLP model. Moreover, the difference in accuracy for the training and test dataset was also higher than the MLP model. Comparison of observed and predicted values in Fig. 4, is a further indication of this trend, wherein the dispersion of values is higher than MLP model.

The accuracy training data shown in Table 5 shows accuracy of GRNN was very close to that of MLP, however, GRNN had the lowest accuracy for the test dataset. This could be an indication of the overfitting of the data. GRNN has also proved to be computationally expensive with more processing layers and hidden neurons in this study. The plot of observed and predicted values, shown in Fig. 6, shows that the values for the test dataset are out of the range of the data. The absolute error for SVM was found to be lowest, as compared to any other technique used in this study. Plot of observed and predicted values, in Fig. 7, also shows an even distribution of data. In another study, GRNN and SVM were compared and SVM was found to be a better prediction tool for predicting DM for high modulus asphalt mixtures59.

The higher performance of MLP, in comparison to other techniques including SVM, is also shown in previous literature for predicting other characteristics of AC mixtures60. The prediction performance of the MLP model, found in this study, was comparable to another previous study61 in which deep learning ANNs were used. Zhang et al.,62 found that with efficient methods of weights optimization in ANN models, their performance can be even better than ensemble learning models.

Based upon the above comparative discussion, it can be concluded that the results of this study conform to the findings of the previous studies in terms of predictive performance of the models. The results and their comparison clearly show the dominance of MLP models for prediction of the DM of asphalt mixtures in different cases. It should be noted that the comparison presented in the current study is more comprehensive, in terms of types of models, than those found in the above-mentioned studies. Furthermore, these studies have not considered the modifications of the LDPE and HDPE in asphalt mixtures, while employing machine learning models. Future research may include calibrating analytical and statistical models for modified AC samples and comparing their accuracy with the machine learning models.

As shown in the previous sub-section, MLP was found to be the most accurate and robust model for predicting the DM of modified AC mixtures in this study. Hence, its weights (Appendix A) were used to establish the importance of input parameters on the model prediction. For this purpose, a relative importance index was used. This index was calculated for each input parameter as per Eq. (1).

Where RIi is the relative importance of each input parameter ‘i’, Wih is the transfer weight for the input parameter to the hidden neuron ‘h’ and Who is the transfer weight for the output of hidden neuron to the output neuron63. There were 4 hidden neurons in the final MLP network while the number of input parameters were 5.

The RI values for the input parameters are shown in Table 6. It can be observed that the frequency had the highest impact on the DM predictions made by the MLP model, followed by similar effects of temperature and inclusion of the LDPE. Figure 3 shows that frequency has a directly proportional impact on the DM, while extreme temperatures greatly reduce the DM values. The LDPE inclusion resulted in the highest DM in all cases. Figure 3 also shows that the modified and unmodified samples showed the same trends at different frequencies and temperatures. Hence, there is no visible difference in the nature of AC mixtures due to the addition of the LDPE and HDPE.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Geyer, R., Jambeck, J. R. & Law, K. L. Production, use, and fate of all plastics ever made. Sci. Adv. 3, 7. https://doi.org/10.1126/sciadv.1700782 (2017).

Article Google Scholar

Ritchie, H. & Roser, M. Plastic pollution. Our World in Data. https://ourworldindata.org/plastic-pollution (2018).

Khan, M. M. A. 4 ways Pakistan is tackling plastic waste and pollution. https://www.weforum.org/agenda/2021/11/4-ways-pakistan-is-tackling-plastic-waste-and-pollution/ (2021).

Khurshid, N. & The Looming Plastic Pollution Crisis. https://www.pakistantoday.com.pk/2019/03/14/the-looming-plastic-pollution-crisis/ (2019).

Plastic Pollution by Country. https://worldpopulationreview.com/country-rankings/plastic-pollution-by-country (2023).

Amin, M. N., Khan, M. I. & Saleem, M. U. Performance evaluation of asphalt modified with municipal wastes for sustainable pavement construction. Sustainability 8(10), 949. https://doi.org/10.3390/su8100949 (2016).

Liang, M. et al. Comparison of rheological properties and compatibility of asphalt modified with various polyethylene. Int. J. Pavement Eng. 22(1), 11–20. https://doi.org/10.1080/10298436.2019.1575968 (2021).

Article Google Scholar

Haider, S., Hafeez, I. & Ullah, R. Sustainable use of waste plastic modifiers to strengthen the adhesion properties of asphalt mixtures. Constr. Build. Mater. 235, 117496. https://doi.org/10.1016/j.conbuildmat.2019.117496 (2020).

Article Google Scholar

Tayfur, S., Ozen, H. & Aksoy, A. Investigation of rutting performance of asphalt mixtures containing polymer modifiers. Constr. Build. Mater. 21(2), 328–337. https://doi.org/10.1016/j.conbuildmat.2005.08.014 (2007).

Article Google Scholar

Rincón-Estepa, J. A., González-Salcedo, E. V., Rondón-Quintana, H. A., Reyes-Lizcano, F. A. & Bastidas-Martínez, J. G. Mechanical behavior of low-density polyethylene waste modified hot mix asphalt. Sustainability 14(7), 4229. https://doi.org/10.3390/su14074229 (2022).

Article Google Scholar

Alshehri, H. A., Wahhab, H. I. A. A., Dalhat, M. A., Al-Juhani, A. H. A. & Al-Osta, M. A. Recycled polyethylene Waste as Binder stabilizer for SMA mix in Gulf Environment. Case Stud. Constr. Mater. 2023, e02177. https://doi.org/10.1016/j.cscm.2023.e02177 (2023).

Nguyen, V. H., Le, V. P. & Nguyen, T. P. Performance evaluation of waste high density polyethylene as a binder modifier for hot mix asphalt. Int. J. Pavement Res. Technol. 2023, 1–12. https://doi.org/10.1007/s42947-023-00331-w (2023).

Ibrahim, A. H. A. Laboratory investigation of aged HDPE-modified asphalt mixes. Int. J. Pavement Res. Technol. 12, 364–369. https://doi.org/10.1007/s42947-019-0043-y (2019).

Article Google Scholar

Mashaan, N., Chegenizadeh, A. & Nikraz, H. A comparison on physical and rheological properties of three different waste plastic-modified bitumen. Recycling 7(2), 18. https://doi.org/10.3390/recycling7020018 (2022).

Article Google Scholar

Eleyedath, A. & Swamy, A. K. Prediction of dynamic modulus of asphalt concrete using hybrid machine learning technique. Int. J. Pavement Eng. 23(6), 2083–2098. https://doi.org/10.1080/10298436.2020.1841191 (2022).

Article Google Scholar

Machsus, M. et al. Improvement for asphalt mixture performance using plastic bottle waste. Geomate J. 20(79), 139–146. https://doi.org/10.21660/2021.79.j2035 (2021).

Article Google Scholar

Goli, A., Rout, B., Cyril, T. & Govindaraj, V. Evaluation of mechanical characteristics and plastic coating efficiency in plastic-modified asphalt mixes. Int. J. Pavement Res. Technol. 16(3), 693–704. https://doi.org/10.1007/s42947-022-00157-y (2023).

Article Google Scholar

Lugeiyamu, L., Kunlin, M., Mensahn, E. S. & Faraz, A. Utilization of waste polyethylene terephthalate (PET) as partial replacement of bitumen in stone mastic asphalt. Constr. Build. Mater. 309, 125176. https://doi.org/10.1016/j.conbuildmat.2021.125176 (2021).

Article Google Scholar

Suksiripattanapong, C. et al. Performance of asphalt concrete pavement reinforced with high-density polyethylene plastic waste. Infrastructures 7(5), 72. https://doi.org/10.3390/infrastructures7050072 (2022).

Article Google Scholar

Cardone, F., Spadoni, S., Ferrotti, G. & Canestrari, F. Asphalt mixture modification with a plastomeric compound containing recycled plastic: Laboratory and field investigation. Mater. Struct. 55(3), 109. https://doi.org/10.1617/s11527-022-01954-4 (2022).

Article Google Scholar

Dalhat, M. A., Al-Abdul Wahhab, H. I. & Al-Adham, K. Recycled plastic waste asphalt concrete via mineral aggregate substitution and binder modification. J. Mater. Civ. Eng. 31, 8 (2019). https://www.researchgate.net/publication/333394962

Article Google Scholar

Vishnu, T. B. & Singh, K. L. A performance study on asphalt concrete mixes with different waste materials as modifiers in pavement application. J. Mater. Cycles Waste Manage. 25(3), 1519–1533. https://doi.org/10.1007/s10163-023-01626-7 (2023).

Article MathSciNet Google Scholar

Benson, A. Predictions of the dynamic complex modulus of non-conventional asphalt concrete using machine learning techniques. https://digitalcommons.georgiasouthern.edu/etd/2593 (2023).

Gong, H. et al. An efficient and robust method for predicting asphalt concrete dynamic modulus. Int. J. Pavement Eng. 23(8), 2565–2576. https://doi.org/10.1080/10298436.2020.1865533 (2022).

Article Google Scholar

Liu, J., Yan, K., Liu, J. & Zhao, X. Using artificial neural networks to predict the dynamic modulus of asphalt mixtures containing recycled asphalt shingles. J. Mater. Civ. Eng. 30(4), 04018051. https://doi.org/10.1061/(ASCE)MT.1943-5533.0002242 (2018).

Article Google Scholar

El-Badawy, S., El-Hakim, A., Awed, A. & R., & Comparing artificial neural networks with regression models for hot-mix asphalt dynamic modulus prediction. J. Mater. Civ. Eng. 30(7), 04018128. https://doi.org/10.1061/(ASCE)MT.1943-5533.0002282 (2018).

Article Google Scholar

Ghasemi, P., Aslani, M., Rollins, D. K. & Williams, R. C. Principal component neural networks for modeling, prediction, and optimization of hot mix asphalt dynamics modulus. Infrastructures 4(3), 53. https://doi.org/10.3390/infrastructures4030053 (2019).

Article Google Scholar

Sakhaeifar, M. S., Underwood, B. S., Kim, Y. R., Puccinelli, J. & Jackson, N. Development of artificial neural network predictive models for populating dynamic moduli of long-term pavement performance sections. Transp. Res. Rec. 2181(1), 88–97. https://doi.org/10.3141/2181-10 (2010).

Article Google Scholar

El-Badawy, S. M., Khattab, A. M., Al, A. A. & Hazmi Using artificial neural networks (ANNs) for hot mix asphalt E* predictions. In Geo-China 2016, 83–91. https://doi.org/10.1061/9780784480076.010 (2016).

Singh, D., Zaman, M. & Commuri, S. Artificial neural network modeling for dynamic modulus of hot mix asphalt using aggregate shape properties. J. Mater. Civ. Eng. 25(1), 54–62. https://doi.org/10.1061/(ASCE)MT.1943-5533.0000548 (2013).

Article Google Scholar

Le, T. H., Nguyen, H. L. & Pham, C. T. Artificial intelligence approach to predict the dynamic modulus of asphalt concrete mixtures. J. Sci. Transp. Technol. 2022, 22–31. https://doi.org/10.58845/jstt.utt.2022.en.2.2.22-31 (2022).

Acharjee, P. K. & Souliman, M. Development of Dynamic Modulus Predictive Model Using Artificial Neural Network (ANN). In 2022 ASEE Gulf Southwest Annual Conference (2022). https://peer.asee.org/39173

Khasawneh, M. A. & Al-Oqaily, D. M. Development of analytical models to predict the dynamic shear rheometer outcome—phase angle. Int. J. Pavement Res. Technol. 16(2), 425–443. https://doi.org/10.1007/s42947-021-00141-y (2023).

Article Google Scholar

Barugahare, J., Amirkhanian, A. N., Xiao, F. & Amirkhanian, S. N. ANN-based dynamic modulus models of asphalt mixtures with similar input variables as Hirsch and Witczak models. Int. J. Pavement Eng. 23(5), 1328–1338. https://doi.org/10.1080/10298436.2020.1799209 (2022).

Article Google Scholar

Lee, D., Derrible, S. & Pereira, F. C. Comparison of four types of artificial neural network and a multinomial logit model for travel mode choice modeling. Transp. Res. Rec. 2672(49), 101–112. https://doi.org/10.1177/0361198118796971 (2018).

Article Google Scholar

Alam, M. S. & Gazder, U. Shear strength prediction of FRP reinforced concrete members using generalized regression neural network. Neural Comput. Appl. 32, 6151–6158. https://doi.org/10.1007/s00521-019-04107-x (2020).

Article Google Scholar

Gopalakrishnan, K. & Kim, S. Support vector machines approach to HMA stiffness prediction. J. Eng. Mech. 137(2), 138–146. https://doi.org/10.1061/(ASCE)EM.1943-7889.0000214 (2011).

Article Google Scholar

Shan, L., Wang, Y., Liu, S., Qi, X. & Wang, J. Establishment of correlation model between compositions and dynamic viscoelastic properties of asphalt binder based on machine learning. Constr. Build. Mater. 364, 129902. https://doi.org/10.1016/j.conbuildmat.2022.129902 (2023).

Article Google Scholar

Lin, M., Hu, C., Easa, S. M. & Jiang, Z. A New Approach to predict dynamic loads considering highway alignment using data mining techniques. Appl. Sci. 12(11), 5719. https://doi.org/10.3390/app12115719 . (2022).

Upadhya, A., Thakur, M. S. & Sihag, P. Predicting Marshall stability of carbon fiber-reinforced asphalt concrete using machine learning techniques. Int. J. Pavement Res. Technol. 2022, 1–21. https://doi.org/10.1007/s42947-022-00223-5 (2022).

Alas, M., Abba, S. I., Ismael Albrka Ali, S., Rahim, A. & Yusoff, N. I. M. Evaluating the performance of aluminum oxide nanoparticle-modified asphalt binder and modelling the viscoelastic properties by using artificial neural networks and support vector machines. Adv. Mater. Sci. Eng.. https://doi.org/10.1155/2022/9685454 (2022).

Upadhya, A., Thakur, M. S., Mashat, A. & Gupta, G. Prediction of binder content in glass fiber reinforced asphalt mix using machine learning techniques. IEEE Access. 10, 33866–33881. https://doi.org/10.1109/ACCESS.2022.3157639 (2022).

Article Google Scholar

Svilar, M., Peško, I. & Šešlija, M. Model for estimating the modulus of elasticity of asphalt layers using machine learning. Appl. Sci. 12, 20. https://doi.org/10.3390/app122010536 (2022).

Article Google Scholar

Institute, A. MS-2 Asphalt Mix Design Methods (Lexington Kentucky, 2014).

Mohammad, L. N., Wu, Z., Obulareddy, S., Cooper, S. & Abadie, C. Permanent deformation analysis of hot-mix asphalt mixtures with simple performance tests and 2002 mechanistic–empirical pavement design software. Transp. Res. Rec. 1970(1), 133–142. https://doi.org/10.1177/03611981061970001 (2006).

Article Google Scholar

Junaid, M., Irfan, M., Ahmed, S. & Ali, Y. Effect of binder grade on performance parameters of asphaltic concrete paving mixtures. Int. J. Pavement Res. Technol. 11(5), 435–444. https://doi.org/10.1016/j.ijprt.2017.11.006 (2018).

Article Google Scholar

Haider, S. et al. Mathematical formulation for predicting moisture damage indices of asphalt mixtures treated with sustainable waste plastic modifiers using gene expression programming. Constr. Build. Mater. 426, 136146. https://doi.org/10.1016/j.conbuildmat.2024.136146 (2024).

Article Google Scholar

Haider, S. & Hafeez, I. A step toward smart city and green transportation: eco-friendly waste PET management to enhance adhesion properties of asphalt mixture. Constr. Build. Mater. 304, 124702. https://doi.org/10.1016/j.conbuildmat.2021.124702 (2021).

Article Google Scholar

Ubbens, J., Cieslak, M., Prusinkiewicz, P. & Ian Stavness. The use of plant models in deep learning: an application to leaf counting in rosette plants. Plant. Methods 14, 1–10 (2018).

Article Google Scholar

Lennox, B., Montague, G. A. & Frith, A. M. Chris Gent, and Vic Bevan. Industrial application of neural networks—an investigation. J. Process Control 11(5), 497–507. https://doi.org/10.1016/S0959-1524(00)00027-5 (2001).

Article Google Scholar

Sapankevych, N. I. & Ravi Sankar. and. Time series prediction using support vector machines: a survey. IEEE Comput. Intell. Mag. 4(2), 24–38. https://doi.org/10.1109/MCI.2009.932254 (2009).

Popescu, M. C., Balas, V. E., Perescu-Popescu, L. & Mastorakis, N. Multilayer perceptron and neural networks. WSEAS Trans. Circuits Syst. 8(7), 579–588 (2009).

Google Scholar

Montazer, G. A., Giveki, D., Karami, M. & Rastegar, H. Radial basis function neural networks: a review. Comput. Rev. J. 1(1), 52–74. https://purkh.com/index.php/tocomp (2018).

Google Scholar

Kulkarni, S. G., Chaudhary, A. K., Nandi, S., Tambe, S. S. & Kulkarni, B. D. Modeling and monitoring of batch processes using principal component analysis (PCA) assisted generalized regression neural networks (GRNN). Biochem. Eng. J. 18(3), 193–210. https://doi.org/10.1016/j.bej.2003.08.009 (2004).

Article Google Scholar

Pisner, D. A. & Schnyer, D. M. Support vector machine. In Machine Learning 101–121 (Academic Press, 2020). https://doi.org/10.1016/B978-0-12-815739-8.00006-7.

Kabiru, O. A., Owolabi, T. O., Ssennoga, T. & Olatunji, S. O. Performance Comparison of SVM and ANN in Predicting Compressive Strength of Concrete (Springer, 2014).

Far, M. S. S., Underwood, B. S., Ranjithan, S. R., Kim, Y. R. & Jackson, N. Application of artificial neural networks for estimating dynamic modulus of asphalt concrete. Transp. Res. Rec. 2127(1), 173–186. https://doi.org/10.3141/2127-20 (2009).

Article Google Scholar

El-Badawy, S., Abd El-Hakim, R. & Awed, A. Comparing artificial neural networks with regression models for hot-mix asphalt dynamic modulus prediction. J. Mater. Civ. Eng. 30, 7. https://doi.org/10.1061/(ASCE)MT.1943-5533.0002282 (2018).

Article Google Scholar

Wang, C. et al. Dynamic Modulus Prediction of a high-Modulus asphalt mixture. Adv. Civ. Eng. 2021, 1–10. https://doi.org/10.1155/2021/9944415 (2021).

Babagoli, R. & Rezaei, M. Development of prediction models for moisture susceptibility of asphalt mixture containing combined SBR, waste CR and ASA using support vector regression and artificial neural network methods. Constr. Build. Mater. 322, 126430. https://doi.org/10.1016/j.conbuildmat.2022.126430 (2022).

Article Google Scholar

Moussa, G. S. & Owais, M. Modeling Hot-Mix asphalt dynamic modulus using deep residual neural networks: Parametric and sensitivity analysis study. Constr. Build. Mater. 294, 123589. https://doi.org/10.1016/j.conbuildmat.2021.123589 (2021).

Article Google Scholar

Zhang, C., Ildefonzo, D. G., Shen, S., Wang, L. & Huang, H. Implementation of ensemble Artificial neural network and MEMS wireless sensors for In-Situ asphalt mixture dynamic modulus prediction. Constr. Build. Mater. 377, 131118. https://doi.org/10.1016/j.conbuildmat.2023.131118 (2023).

Article Google Scholar

Kemp, S. J., Zaradic, P. & Hansen, F. An approach for determining relative input parameter importance and significance in artificial neural networks. Ecol. Model. 204(3–4), 326–334. https://doi.org/10.1016/j.ecolmodel.2007.01.009 (2007).

Download references

The authors express their gratitude to Southwest Jiaotong University (International Postgraduate Education and Teaching Reform Research Project Grand No. GYJG [2023] Y13) for providing the necessary facilities and resources to support the experimental work in this study.

School of Transportation and Logistics, Southwest Jiaotong University (SWJTU), Chengdu, China

Muhammad Junaid & Chaozhe Jiang

Department of Civil Engineering, University of Bahrain, Zallaq, Bahrain

Uneb Gazder

Department of Civil Engineering, University of Engineering and Technology, Taxila, Pakistan

Imran Hafeez

Doctoral School, Silesian University of Technology, Akademicka 2a, Gliwice, 44-100, Poland

Diyar Khan

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

Conceptualization, Muhammad Junaid., and Chaozhe Jiang.; methodology, Muhammad Junaid., and Diyar Khan.; software and formal analysis, and Muhammad Junaid; data curation, Diyar Khan, Uneb Gazder and Chaozhe Jiang; writing—original draft preparation, Muhammad Junaid.; writing—review and editing, Chaozhe Jiang, Diyar Khan, Imran Hafeez and Uneb Gazder, Machine learning (ML); Muhammad Junaid, Chaozhe Jiang, Diyar Khan and Uneb Gazder. All authors have read and agreed to the published version of the manuscript.

Correspondence to Chaozhe Jiang, Uneb Gazder or Diyar Khan.

The authors declare no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Below is the link to the electronic supplementary material.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

Junaid, M., Jiang, C., Gazder, U. et al. Evaluation of asphalt mixtures modified with low-density polyethylene and high-density polyethylene using experimental results and machine learning models. Sci Rep 14, 24601 (2024). https://doi.org/10.1038/s41598-024-74657-1

Download citation

Received: 01 August 2024

Accepted: 27 September 2024

Published: 19 October 2024

DOI: https://doi.org/10.1038/s41598-024-74657-1

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative