Solar Radiation ML Models – Saudi Arabian Dataset

This repository contains the full machine learning workflow to predict Global Horizontal Irradiance (GHI) using Saudi Arabia’s weather data (2015–2020).

Objectives

• Train & compare multiple ML models (Linear Regression, RF, XGB, etc.)
• Evaluate with MAE, RMSE, R²
• Export best model for deployment

Models Evaluated

• Linear Regression (used for deployment)
• Random Forest
• XGBoost
• Histogram Gradient Boosting
• Support Vector Regression
• Artificial Neural Networks
• Decision Tree
• KNN

Model Selection Strategy

Multiple regression models were evaluated and tracked using Weights & Biases (W&B), including Linear Regression, Random Forest, Decision Tree, KNN, SVR, XGBoost, ANN, and Histogram Gradient Boosting.

Models were automatically ranked using a composite score that balances:

• Prediction accuracy (RMSE, R²)
• Inference latency (testing time) Although some complex models achieved comparable accuracy, Linear Regression provided the best accuracy–latency trade-off and was therefore selected for deployment.

Full experiment tracking and interactive dashboards are available on: https://wandb.ai/nishnarudkar-d-y-patil-university/solar-radiation-prediction

Weights and Biases Report: https://api.wandb.ai/links/nishnarudkar-d-y-patil-university/l4xo0wax

Contents

• notebooks/: Model training and evaluation
• models/: Saved .pkl files
• dataset/: Cleaned sample dataset
• requirements.txt: Library dependencies

Highlights

• Best R²: 0.97 (Linear Regression)
• 21 input features including DHI, DNI, humidity, wind speed
• 10-fold and 43-fold cross-validation

Live App (Deployed Model)

See deployment repo here: (https://solar-radiation-prediction-using-saudi.onrender.com)

Related Repositories:

• Model Training & Evaluation Repo
• Web App Deployment Repo