FS-GP

This repository provides the implementation for the study:

"Benchmarking Feature Selection Methods and Prediction Models for Flowering Time Prediction in Plant Breeding" (IJMS)

The code supports full reproducibility of all experiments reported in the manuscript.

Contents

This repository includes:

1. Fold generation and nested cross-validation splits
2. Feature selection (FS) methods
3. Model training and hyperparameter tuning
4. Model evaluation and performance analysis
5. SHAP-based feature importance visualization

• FS/feature_selectors.py
  Implements all feature selection methods, including: ElasticNet, LASSO, Random Forest, XGBoost, LightGBM, Mutual Information, and Boruta.

• FS/Prediction.py
  Main script for running the nested cross-validation pipeline. It calls feature selection methods from feature_selectors.py, performs model training, hyperparameter tuning, and evaluation.

• FS/RRBLUP.R
  Script for genomic prediction using RRBLUP with exported training/testing splits.

• SHAP/SHAP.py
  Computes and visualizes feature importance using SHAP values.

• SHAP/RF_model_Boruta.pkl
  Pre-trained Random Forest model using Boruta-selected features, provided to directly reproduce the SHAP analyses and prediction results.