MChalwa/Ghana-Rainfall-Prediction

This project builds an end-to-end machine learning pipeline to predict rainfall intensity (No Rain, Small Rain, Medium Rain, Heavy Rain) using meteorological indicators. # Ghana-Rainfall-Prediction This project builds an end-to-end machine learning pipeline to predict rainfall intensity (No Rain, Small Rain, Medium Rain, Heavy Rain) using meteorological indicators. ## Project Overview This project develops an end-to-end machine learning pipeline to classify rainfall intensity into four categories: - NORAIN - SMALLRAIN - MEDIUMRAIN - HEAVYRAIN Using historical meteorological data, the model learns patterns from indicators such as dew, fog, wind, clouds, heat, lightning, and more. The project is part of a broader initiative to improve early rainfall forecasting and support climate-related decision-making. ## Key Features 1. Cleaned and preprocessed meteorological data 2. Filled missing indicator values based on domain-driven mapping 3. Built a unified preprocessing and modeling pipeline using ColumnTransformer + Pipeline 4. Compared multiple machine learning algorithms: - Logistic Regression - Decision Tree - Random Forest - SVM - XGBoost - LightGBM - CatBoost CatBoost achieved the best performance (F1-score: 0.95) 5. Generated predictions for the test dataset 6. Prepared final id and Target submission file ## Modeling Approach 1. Data Preprocessing Handled missing values Encoded the target variable Applied One-Hot Encoding to categorical features Scaled numeric features 2. Model Training & Evaluation Evaluated multiple models on the validation set using F1-score. CatBoost performed best due to its ability to handle categorical data efficiently. 3. Final Prediction The best model was retrained on the full training data Predictions were generated for the unseen test dataset Output saved as a submission file ## Conclusions 1. Machine learning techniques can effectively predict rainfall intensity in Ghana, with several models achieving high accuracy and strong F1-scores (>95%) on the validation set. - CatBoost, XGBoost, and LightGBM performed the best, indicating that gradient boosting algorithms handle mi …