Shikha2u/waterpump-failure-imbalanced-classification

End-to-end ML on 59K+ Tanzania water points: EDA, feature engineering, and 3-class pump status prediction # Tanzania Water Pump Status Prediction — EDA, Feature Engineering & Multi-Class ML > **AquaPredict** · End-to-end machine learning pipeline on 59K+ Tanzania water points: data cleaning, exploratory analysis, multi-class classification (functional / needs repair / non-functional), and model evaluation using DrivenData Pump It Up data. ## Impact Millions of people in Tanzania depend on water pumps for clean drinking water. Predicting which pumps are functional, need repair, or are non-functional helps prioritize maintenance and improve rural water access. This project builds a full classification pipeline on real open data. ## About the Data Dataset: **DrivenData Pump It Up** — Taarifa open data on Tanzania water points (~59,000 training records). **Target:** 3-class classification - `functional` - `functional needs repair` - `non functional` ## Key Results - End-to-end pipeline: data cleaning → EDA → feature engineering → classification - Exploratory analysis of geographic, operational, and construction features - Multi-class model with evaluation metrics ## Tech Stack Python · pandas · scikit-learn · Jupyter · matplotlib · seaborn ## Documentation | Doc | Description | |-----|-------------| | Problem Statement | Context and objectives | | Data Dictionary | Key features and columns | | Modeling Approach | Pipeline overview | | Final Report | Complete analysis write-up | ## Project Structure ``` waterpump-failure-imbalanced-classification/ ├── notebooks/ │ ├── 01_eda_and_exploration.ipynb │ └── 02_classification_model.ipynb ├── src/ │ └── clean_water_data.py ├── data/ │ ├── training_values.csv │ └── training_labels.csv ├── docs/ └── reports/ ``` ## Setup & Usage ```bash pip install -r requirements.txt jupyter notebook notebooks/01_eda_and_exploration.ipynb ``` To regenerate cleaned data: ```bash python src/clean_water_data.py ``` ## Skills Demonstrated - Exploratory data …