Military_Models – Complete Retraining SVR Model (Approach 3)

Model Scripts created in Python JupyterLab
Author: Ayden McCarthy
Manuscript Title: Optimisation of a Support Vector Regression Model Predicts Individuals’ Maximal Weight Lifted During a Simulated Military Manual Material Handling Assessment
Program of Study: PhD
Institution: Macquarie University
Year: 2025

---

📘 Overview

This repository contains the finalized Support Vector Regression (SVR) model developed using the complete retraining approach (Approach 3). The model predicts maximal weight lifted during a simulated military manual material handling task. It was built using both general and military datasets and supports full model retraining, feature selection, evaluation, and deployment.

Included files:

• SVR_App_3.ipynb – Notebook to train and evaluate the model with SHAP analysis and SBS feature selection
• predict_complete_retrained_model.ipynb – Notebook to predict new cases using the saved model
• svr_complete_retrained_model.joblib – Saved model and scaler for reuse

---

🧠 Model Summary

• Model: Support Vector Regression (SVR)
• Kernel: Sigmoid (selected via GridSearchCV)
• Feature Selection: Sequential Backward Selection (SBS)
• Evaluation: Leave-One-Out Cross-Validation (LOO)
• Explained with: SHAP values

Saved model includes both the SVR regressor and the StandardScaler used for feature normalization.

---

🛠️ Training Instructions

To train and evaluate the complete retrained SVR model:

1. Place the following datasets in your working directory:

   • Training_Set_Gen_Pop_No_MP.csv
   • Training_Set_Mil_Pop_No_MP.csv

2. Open and run all cells in SVR_App_3.ipynb. This notebook performs:

   • Data loading and concatenation
   • Grid search with LOO CV to optimize SVR hyperparameters
   • Sequential backward feature selection (SBS)
   • SHAP analysis for model interpretability
   • Final model evaluation and prediction visualization

Optional test set evaluation with:

• Testing_Set_Gen_Pop_No_MP.csv
• Testing_Set_Mil_Pop_No_MP.csv

---

🗂️ Input Format for Prediction

For new predictions, prepare a CSV file named new_data.csv with the following columns:

• IMTP_Peak Vertical Force [N]
• Avg_Bicep
• LCMJ_Velocity at Peak Power [m/s]
• Concentric Peak Velocity [m/s]
• Peak Power [W]
• Sex
• Height (cm)
• Weight (kg)
• Age

⚠️ Do NOT include the target Lift-to-Place. This is the model output.

---

🔍 Prediction Workflow

To generate predictions on new data:

1. Place these in your working directory:

   • svr_complete_retrained_model.joblib
   • new_data.csv

2. Run the notebook:

   • predict_complete_retrained_model.ipynb

This script will:

• Auto-install any missing packages (pandas, joblib, etc.)
• Load the trained SVR model and associated scaler
• Apply scaling and generate predictions

Example Output:

Predicted Lift-to-Place value(s): [78.45 83.12 69.30]

---

⚙️ Environment Setup

To install required packages manually (see requirements.txt):

pip install pandas numpy scikit-learn matplotlib joblib shap mlxtend

Or let the notebooks install them automatically on first run.

---

📜 License

This project is licensed under the MIT License. You are free to use, adapt, and redistribute the code with appropriate attribution.

---

🙏 Citation

If you use or refer to this model in your work, please cite:

McCarthy, A. (2025). Optimisation of a Support Vector Regression Model Predicts Individuals’ Maximal Weight Lifted During a Simulated Military Manual Material Handling Assessment. PhD Thesis, Macquarie University.

---

🤝 Issues & Contributions

Questions, improvements, and collaborations are welcome. Please open a GitHub issue or submit a pull request.