ANN‑stacked hybrid model of the van Genuchten and Fredlund–Xing models for improved prediction of matric suction in a silty clay soil

ANN Hybrid SWCC Modeling – Dataset, Codes, and Trained Models This repository contains all datasets, code files, and trained Artificial Neural Network (ANN) models used for the development of a hybrid Soil-Water Characteristic Curve (SWCC) prediction framework that integrates Van Genuchten (VG), Fredlund–Xing (FX), and ANN models. The files are organized into three main folders: 1. ANN_models/ This folder contains the trained ANN models and the corresponding data scalers used during model training. Files include: ann_model_600.h5 – ANN model trained using 600-row generated dataset ann_model_3000.h5 – ANN model trained using 3000-row generated dataset ann_model_9000.h5 – ANN model trained using 9000-row generated dataset Scalers (used to normalize/denormalize inputs and outputs): scaler_X_600.pkl, scaler_y_600.pkl scaler_X_3000.pkl, scaler_y_3000.pkl scaler_X_9000.pkl, scaler_y_9000.pkl These files are required to correctly preprocess inputs before prediction and to convert model outputs back to physical units. 2. Codes/ This folder contains all code used for: Data generation ANN training Model evaluation Plot generation Hybrid SWCC prediction and comparison with VG and FX models File included: ANN Hybrid SWCC modeling.ipynb – Jupyter Notebook containing the full workflow: Synthetic data generation ANN hyperparameter tuning Training best-performing models Evaluating models on experimental suction–water content data Plotting ANN vs VG vs FX curves 3. Suction_data/ This folder contains the raw and generated datasets used for training and validation. Files include: 600_generated_data.csv – Generated SWCC dataset (600 points) 3000_generated_data.csv – Generated SWCC dataset (3000 points) 9000_generated_data.csv – Generated SWCC dataset (9000 points) Experimental Suction data.xlsx – Measured experimental suction–water content data used for evaluation These datasets include suction (ψ, kPa) and volumetric water content (θ) pairs with noise-added synthetic samples for ANN training. How to Use the Repository Open the Jupyter Notebook in /Codes/ Load the appropriate model and scalers from /ANN_models/ Load experimental or custom suction data from /Suction_data/ Run predictions or regenerate the plots Compare ANN predictions with VG and FX models Contact For questions or clarification, please contact the author.