TY  - JOUR
AU  - Gnananandarao, Tammineni
AU  - Muktinutalapati, Jayatheja
AU  - Maralapalle, Vedprakash
AU  - Kumar, B. A. V. Ram
AU  - Ajay, CH.
PY  - 2025
DA  - 2025/09/24
TI  - Advanced Machine Learning and Optimization for Erodibility Prediction of Treated Unsaturated Lateritic Soil
JO  - Sustainable Intelligent Infrastructure
T2  - Sustainable Intelligent Infrastructure
JF  - Sustainable Intelligent Infrastructure
VL  - 1
IS  - 2
SP  - 93
EP  - 107
DO  - 10.62762/SII.2025.839324
UR  - https://www.icck.org/article/abs/SII.2025.839324
KW  - unsaturated lateritic soil
KW  - ANN
KW  - RF
KW  - SVR
KW  - XGBoost
KW  - CatBoost
KW  - PSO
KW  - relative importance
AB  - This research pioneers the application of a diverse set of advanced machine learning and optimization methods, to predict the erodibility of lateritic soil treated with cement and nanostructured quarry fines, providing a groundbreaking, data-driven approach that enhances traditional erosion analysis techniques. Traditional experimental methods for erosion analysis are often complex and resource-intensive; therefore, this research focuses on developing predictive models using Python. To build the machine learning and optimization models, 121 data points were collected from existing literature. The dataset includes erodibility measurements of unsaturated lateritic soil treated with local cement and enhanced with nanostructured quarry fines. The study employs Artificial Neural Networks (ANN), Random Forest (RF), Support Vector Machine (SVM), XGBoost, CatBoost, and Particle Swarm Optimization (PSO) to predict soil erodibility. The data was divided into training (70%), testing (15%), and validation (15%) sets for model development and evaluation. Model performance was assessed using statistical metrics such as $R^2$, M.A.E., M.S.E., R.M.S.E., and M.A.P.E. The results indicated an $R^2$ value are almost equal to 1 in training, testing, and validation phases, and the M.A.P.E. values are below 3% for the CatBoost, RF, XGB, SVM, and ANN models across all three phases: training, testing, and validation. The CatBoost, RF, XGB, SVM, and ANN models are most accurate in predicting the erodibility. Finally, relative importance showed that maximum unit weight and hydrated cement are the most influencing parameter in predicting the erodibility.
SN  - 3067-8137
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Gnananandarao2025Advanced,
  author = {Tammineni Gnananandarao and Jayatheja Muktinutalapati and Vedprakash Maralapalle and B. A. V. Ram Kumar and CH. Ajay},
  title = {Advanced Machine Learning and Optimization for Erodibility Prediction of Treated Unsaturated Lateritic Soil},
  journal = {Sustainable Intelligent Infrastructure},
  year = {2025},
  volume = {1},
  number = {2},
  pages = {93-107},
  doi = {10.62762/SII.2025.839324},
  url = {https://www.icck.org/article/abs/SII.2025.839324},
  abstract = {This research pioneers the application of a diverse set of advanced machine learning and optimization methods, to predict the erodibility of lateritic soil treated with cement and nanostructured quarry fines, providing a groundbreaking, data-driven approach that enhances traditional erosion analysis techniques. Traditional experimental methods for erosion analysis are often complex and resource-intensive; therefore, this research focuses on developing predictive models using Python. To build the machine learning and optimization models, 121 data points were collected from existing literature. The dataset includes erodibility measurements of unsaturated lateritic soil treated with local cement and enhanced with nanostructured quarry fines. The study employs Artificial Neural Networks (ANN), Random Forest (RF), Support Vector Machine (SVM), XGBoost, CatBoost, and Particle Swarm Optimization (PSO) to predict soil erodibility. The data was divided into training (70\%), testing (15\%), and validation (15\%) sets for model development and evaluation. Model performance was assessed using statistical metrics such as \$R^2\$, M.A.E., M.S.E., R.M.S.E., and M.A.P.E. The results indicated an \$R^2\$ value are almost equal to 1 in training, testing, and validation phases, and the M.A.P.E. values are below 3\% for the CatBoost, RF, XGB, SVM, and ANN models across all three phases: training, testing, and validation. The CatBoost, RF, XGB, SVM, and ANN models are most accurate in predicting the erodibility. Finally, relative importance showed that maximum unit weight and hydrated cement are the most influencing parameter in predicting the erodibility.},
  keywords = {unsaturated lateritic soil, ANN, RF, SVR, XGBoost, CatBoost, PSO, relative importance},
  issn = {3067-8137},
  publisher = {Institute of Central Computation and Knowledge}
}