WeightlossPredictionModel

Random Forest Model for Weight Loss Prediction

This project implements a Random Forest classifier to predict weight loss outcomes based on demographic, psychological, and physiological features. The workflow includes data preprocessing, model tuning using Bayesian optimization, and explainability analysis with SHAP.

Features

• Data Preprocessing: Splits data into train, validation, and test sets with stratification based on gender and ethnicity.
• Model Tuning: Utilizes BayesSearchCV to optimize hyperparameters of the Random Forest classifier.
• Model Evaluation: Calculates metrics such as F1 score, sensitivity, specificity, ROC AUC, and precision-recall.
• Explainability: Uses SHAP for feature importance and interpretability of predictions.
• Threshold Optimization: Determines the best decision threshold for the classifier to maximize F1 score.

Dependencies

Key Python packages used:

• pandas
• numpy
• scipy
• scikit-learn
• shap
• matplotlib
• skopt

Usage

1. Preprocess the dataset (SMART_Trial_Weight_Features.csv) for model training.
2. Tune the Random Forest hyperparameters using Bayesian optimization.
3. Evaluate the model on test data and visualize results.
4. Use SHAP to analyze feature contributions to predictions.

Output

• Classification metrics (e.g., F1 scores, ROC AUC).
• SHAP summary plots and force plots.
• Optimized decision threshold for binary classification.