JasmineElmahalawy/Pump-it-Up-Data-Mining-the-Water-Table

A machine learning project for classifying water pump functionality in Tanzania using various classification algorithms. # Pump it Up: Data Mining the Water Table A machine learning project for classifying water pump functionality in Tanzania using various classification algorithms. This project was part of the CC519 Data Mining Course at the AASTMT. ## 🏆 Competition This project was developed for the "Pump it Up: Data Mining the Water Table" competition on DrivenData, focusing on improving water access through predictive analytics. ## Project Overview This project tackles the challenge of predicting the operational status of water pumps in Tanzania using machine learning techniques. The goal is to classify pumps into three categories: - **Functional**: Working properly - **Functional needs repair**: Working but requires maintenance - **Non-functional**: Not working ## Results We achieved **80.98% accuracy** with our best-performing model, demonstrating the potential for machine learning in predictive maintenance of water infrastructure. ### Model Performance Comparison | Model | Accuracy | Best For | |-------|----------|----------| | **Random Forest** | **81%** | Overall best performance with balanced metrics | | XGBoost | 80% | Good overall performance | | CatBoost | 80% | Highest precision for "needs repair" category | | Decision Tree | 74% | Baseline model | ## Technical Approach ### Models Implemented - **Decision Tree**: Simple baseline model - **Random Forest**: Ensemble method with best overall performance - **XGBoost**: Gradient boosting algorithm - **CatBoost**: Advanced gradient boosting for categorical features ### Dataset - **Source**: Taarifa water pump dataset from DrivenData competition - **Size**: 59,000+ water points in Tanzania - **Features**: Geographic location, construction year, water quality, pump type, operational status - **Split**: 80% training, 20% testing ### Evaluation Metrics - Accuracy - Precision, Recall, and F1-Score for each class - Balanced performance across all three categories ## Key Findings 1. **Random Forest emerged as the top pe …