4ra-BB/tanzania-water-pumps-prediction

Machine learning pipeline to predict water pump functionality in Tanzania (~59K pumps), supporting maintenance prioritization for NGOs and government agencies. Random Forest with correlation-based imputation and feature selection. Built as part of the DrivenData "Pump it Up" competition. # Predicting Water Pump Functionality in Tanzania Predictive classification of water pump operational status using machine learning, applied to infrastructure maintenance in sub-Saharan Africa. ## The problem Tanzania has tens of thousands of water pumps distributed across the country. Many fall into disrepair without timely maintenance, leaving communities without access to clean water. Predicting which pumps are likely to fail can help governments and NGOs allocate maintenance resources more effectively. ## Dataset Data from the DrivenData competition "Pump it Up: Data Mining the Water Table", containing information on ~59,400 water pumps. Each pump is classified as: - **functional** : the pump is operational - **functional needs repair** : the pump works but needs maintenance - **non functional** : the pump is not working ## Approach | Step | Method | |---|---| | Imputation | Correlation-based: each variable imputed using the most related variable as grouping key | | Feature selection | Random Forest importance scores, aggregated by original variable | | Model | Random Forest (100 estimators) | | Key finding | Minimal variable recoding outperformed heavy feature engineering | ## Repository structure ``` ├── README.md ├── data/ │ ├── TrainingValues.csv │ ├── TrainingLabels.csv │ └── TestValues.csv ├── notebooks/ │ └── water_pump_status_prediction.ipynb └── .gitignore ``` ## How to run ```bash git clone github.com cd tanzania-water-pumps-prediction pip install pandas numpy scikit-learn matplotlib seaborn scipy jupyter notebook notebooks/water_pump_status_prediction.ipynb ``` ## Tools Python · pandas · scikit-learn · matplotlib · seaborn · scipy ## Author **Laura Benkel Brander** — Sociologist & Data Scientist LinkedIn