Skip to main content
Have a personal or library account? Click to login
Predicting Dynamic Modulus of Asphalt Mixtures Using Explainable Artificial Intelligence Techniques: A Design Optimization Application Cover

Predicting Dynamic Modulus of Asphalt Mixtures Using Explainable Artificial Intelligence Techniques: A Design Optimization Application

Open Access
|Nov 2026
Read online

Figures & tables

Figure 1:

Experimental program to acquire data for predicting dynamic modulus

Table 1:

Key properties of aggregates

Aggregate PropertyReference StandardAverage Value [%]Value Range [%]
SoundnessAASHTO T1043≤12
Sand Equivalent ValueAASHTO T 17672≥ 45
Elongation RatioASTM D 47917≤ 10
Flakiness RatioASTM D 47915≤ 10
Los Angeles AbrasionAASHTO T9626≤ 30
Table 2:

Gradation blend details

Gradation BlendSieve Size [mm]/Percent Passing [%]
37.525.019.012.59.54.752.360.300.075
GB-1---82.568.749.330.810.55.2
GB-2--99.979.569.448.735.311.85.2
GB-3--99.784.563.446.030.510.35.0
GB-4---80.067.048.033.011.04.7
GB-5-82.669.759.252.241.624.59.45.0
GB-6-85.471.059.953.435.526.410.84.7
GB-798.882.767.456.243.535.025.17.64.2
GB-8--95.076.063.042.529.08.55.0
GB-9---82.070.050.030.010.05.0
GB-10---94.087.065.037.09.05.0
GB-11---95.082.059.043.013.06.0
GB-1295.077.065.552.544.031.522.57.04.5
G-13-82.572.562.552.537.525.010.04.5
GB-14-95.085.060.047.030.020.08.04.0
GB-15-95.083.070.063.052.039.014.05.5
GB-16--95.083.065.046.030.09.05.0
Table 3:

Key properties of bitumen

Bitumen PropertyReference StandardARL 40/50ARL 80/100ARL 60/70NRL 60/70
Penetration Value [0.1 mm]AASHTO T4943846664
Softening Point [°C]AASHTO T53524648.849.2
Flash and Fire Points [°C]AASHTO T48335 & 359233 & 275304 & 334310 & 343
Viscosity at 135°C [ Pa·s]AASHTO T3160.4860.2260.2410.292
Table 4:

Summary of mix design parameters using the Marshall method

Bitumen Source/TypeGradation BlendAir Voids [%]Optimum Binder Content [%]Voids in Mineral Aggregate [%]Voids Filled with Asphalt [%]Stability [Kg]Flow [mm]
ARL 40/50G-84.03.913.0969.4411982.29
G-104.05.515.2273.7114472.63
G-144.03.413.1669.6222433.40
G-154.04.413.5570.4831149.00
ARL 80/100G-84.03.711.3264.667902.68
G-104.04.615.0473.409342.58
G-144.03.311.4565.0725268.81
G-154.04.312.3067.506107.52
ARL 60/70G-34.04.214.2066.5013433.10
G-44.04.213.4059.5012982.99
G-74.03.512.8058.0026325.02
G-84.04.012.1566.0813623.00
G-94.04.112.9565.1612913.16
G-104.05.014.7069.1814243.38
G-114.04.814.5267.7315443.28
G-124.03.311.3862.1926504.81
G-134.03.711.8763.3329054.66
G-144.03.612.1761.9322955.38
G-154.03.912.4061.3334964.53
G-164.04.3614.4374.5332359.12
NRL 60/70G-14.0414.866.912902.88
G-24.04.415.363.513303.052
G-54.03.313.758.128904.302
G-64.03.413.259.226204.981
Figure 2:

Representative HMA specimens at different bitumen contents

Figure 3:

Core and uncore specimens for dynamic modulus testing

Figure 4:

Conducting SPT on a cored sample

Table 5:

Model configurations and hyperparameters used in the MATLAB Regression Learner application

ModelMATLAB functionModel specificationMain settings
LRfitlmLinear regressionRobustOpts = off
ILRfitlmInteraction regressionRobustOpts = off
FTfitrtreeFine regression treeMinLeafSize = 4; Surrogate = off
MTfitrtreeMedium regression treeMinLeafSize = 12; Surrogate = off
QSVMfitrsvmQuadratic polynomial SVMPolynomialOrder = 2; KernelScale = auto; BoxConstraint = IQR/1.349; Epsilon = IQR/13.49; Standardize = true
CSVMfitrsvmCubic polynomial SVMPolynomialOrder = 3; KernelScale = auto; BoxConstraint = IQR/1.349; Epsilon = IQR/13.49; Standardize = true
EBTfitrensembleLSBoost30 learning cycles; LearnRate = 0.1; tree MinLeafSize = 8; variables sampled = all
EBagTfitrensembleBagging30 learning cycles; tree MinLeafSize = 8; variables sampled = all
Figure 5:

Histogram matrix showing the frequency distributions of key material and rheological parameters

Figure 6:

Spearman Correlation matrix heatmap showing relationships among mixture components

Table 6:

Metric analysis results for all models

ModelRMSE [psi] (Validation)MSE [psi2] (Validation)R2 [-] (Validation)MAE [psi] (Validation)MAE [psi] (Test)MSE [psi2] (Test)RMSE [psi] (Test)R2 [-] (Test)
LR3.87 × 1051.50 × 10110.863.00 × 1052.93 × 1051.32 × 10113.64 × 1050.90
ILR3.22 × 1051.03 × 10110.902.32 × 1052.05 × 1057.78 × 10102.78 × 1050.94
FT3.97 × 1051.58 × 10110.853.11 × 1052.86 × 1051.49 × 10113.86 × 1050.89
MT4.37 × 1051.91 × 10110.823.38 × 1052.86 1051.46 × 10113.82 × 1050.89
QSVM2.78 × 1057.73 × 10100.931.96 × 1051.80 × 1056.18 × 10102.48 × 1050.95
CSVM2.32 × 1055.39 × 10100.951.67 × 1051.51 × 1054.90 × 10102.21 × 1050.96
EBT3.11 × 1059.67 × 10100.912.40 × 1052.08 × 1057.99 × 10102.82 × 1050.94
EBagT3.42 × 1051.17 × 10110.892.55 × 1052.21 × 1058.73 × 10102.95 × 1050.93
Figure 7:

Actual vs predicted dynamic modulus (Psi) a) LT b) ILT c) FT d) MT e) QSVM f) CSVM g) EBT h) EBagT

Table 7:

Sensitivity Analysis for phase angle

ModelPhase Angle RTest2 [-]No-Phase Angle RTest2 [-]Phase Angle RMSE [psi]No-Phase Angle RMSE [psi]
LR0.900.07043.64 × 1051.03 x 106
ILR0.940.06362.78 × 1051.03 x 106
FT0.89−0.21803.86 × 1051.17 x 106
MT0.89−0.08553.82 × 1051.11 x 106
QSVM0.95−0.02192.48 × 1051.07 x 106
CSVM0.96−0.18542.21 × 1051.16 x 106
EBT0.940.01502.82 × 1051.05 x 106
EBagT0.930.02622.95 × 1051.05 x 106
Figure 8:

Shapley feature importance analysis for Model [Cubic SVM]

Figure 9:

SHAP summary plot for the Cubic SVM model

Figure 10:

SHAP dependence plots for the Cubic SVM model

DOI: https://doi.org/10.2478/cee-2027-0032 | Journal eISSN: 2199-6512 (formerly 1336-5835) | Journal ISSN: 1336-5835
Language: English
Submitted on: Aug 8, 2026
Accepted on: Oct 7, 2026
Published on: Nov 18, 2026
Published by: University of Žilina
In partnership with: Paradigm Publishing Services
Publication frequency: 4 issues per year

© 2026 Muhammad Ajmal, Uneb Gazder, Muhammad Junaid, Rashed Ismaeel, published by University of Žilina
This work is licensed under the Creative Commons Attribution 4.0 License.