Richecard-Blade/WELLS-_CONDITIONPREDICTION-IN-TANZANIA

# Water Wells Condition Prediction in Tanzania ## Project Overview Access to safe water is a critical issue in Tanzania, where millions of people rely on water wells. Some wells are fully functional, some need maintenance, and others are completely non-functional. The objective of this project is to **predict the condition of water wells** using historical data. By accurately classifying wells into three categories—**Functional**, **Functional Needs Repair**, and **Non-Functional**—we aim to help stakeholders prioritize maintenance and improve water accessibility efficiently. --- ## Dataset The dataset contains information about the wells, including: - Pump type and source - Installation year - Geographical coordinates - Basin and region - Technical characteristics like well depth and water quality The target variable is `status_group`, indicating the condition of each well. The dataset also includes a test set for which we provide predictions. --- ## Data Preparation - Missing values were handled: - **Numerical features**: imputed with the median - **Categorical features**: imputed with the most frequent value - Only **numerical features** were used for the final model as requested. - Feature engineering included: - Well age = `recorded_year - construction_year` - Extracted year and month from `date_recorded` - Numerical features were scaled when necessary for models like Logistic Regression. --- ## Modeling Two models were trained and compared: 1. **Logistic Regression (Baseline)** - Multinomial classification - Class weighting to address imbalance - Validation results: - Macro-F1: 0.403 - Accuracy: 47% - Recall for non-functional wells: 54% 2. **Random Forest (Tuned)** - Hyperparameters optimized for maximum performance - Validation results: - Macro-F1: 0.613 - Accuracy: 70% - Recall for non-functional wells: 70% **Selected Model:** Random Forest, due to its superior ability to identify **non-functional wells**, which is the most critical category for …