GitHub | LinkedIn | Twitter

Introduction

Access to clean and potable water is a fundamental necessity, yet many regions, including Tanzania, face challenges in providing this essential resource. The IHH Humanitarian Relief Foundation, an NGO dedicated to improving water access, strives to efficiently allocate their maintenance efforts by accurately predicting the functionality of water pumps. By building a classification model, they can optimize their operations, maximize maintenance resources, and ensure clean water is readily available to the people of Tanzania.

The Cost of Errors

With Tanzania's vast geographical area and limited resources, it is crucial to deploy maintenance and repair efforts judiciously. The country's infrastructure spans over 21,000 miles of roadways, making targeted repairs essential. Constructing a well in Tanzania can cost upwards of $10,000, considering factors such as labor, drilling depth, rock density, location, and fuel costs. Repairing wells, which can range from a few hundred to several thousand dollars, is an expense that should be allocated only to wells in genuine need.

Baseline Model and Simple Model Performance

The baseline model, which predicts all wells as functional, achieved an accuracy of approximately 54%. However, this approach proved insufficient for accurate predictions. Consequently, several simple models were explored, including Logistic Regression, Decision Tree Classifier, Random Forest Classifier, Gradient Booster Model, and XGBoost Model.

Among these models, Logistic Regression emerged as the optimal choice. It achieved an accuracy of approximately 79.1%, outperforming the other models in terms of speed, interpretability, and resistance to overfitting. The Logistic Regression model was further refined using GridSearchCV to identify the best hyperparameters, including the mean imputation strategy for numerical values, a C value of 1.0, penalty 'l2', and the 'liblinear' solver.

The Final Logistic Regression Model

Considering the similar performance of multiple models, Logistic Regression was selected as the final model due to its efficiency in training, interpretability, and robustness against overfitting. The final Logistic Regression model achieved an accuracy of approximately 79.6%. This model provides IHH with a reliable tool for predicting the functionality of water pumps, aiding in prioritizing maintenance efforts and optimizing resource allocation.

Further Exploration and Questions

While the binary classification of functional and non-functional wells is valuable, there is potential benefit in identifying specific wells that require repairs while still being functional. Such insight could enable targeted preventive maintenance, avoiding costly repairs in the future.

Given more time and resources, it would be worthwhile to create a model that predicts the original status groups ('functional,' 'non-functional,' and 'functional needs repair') instead of converting the target into a binary outcome. This expanded model could provide more detailed information and enhance decision-making processes.

Understanding the limiting factors in delivering resources to wells that require repairs is vital. Identifying the challenges related to maintenance professionals, time constraints, financial resources, availability of parts, and knowledge gaps will help IHH devise effective strategies for addressing these obstacles.

Conclusion

The implementation of a logistic regression model has empowered the IHH Humanitarian Relief Foundation to enhance their water well maintenance operations in Tanzania. By accurately predicting the functionality of water pumps, the NGO can efficiently allocate resources and prioritize repairs where they are most needed. As ongoing efforts continue to optimize resource utilization, explore more granular predictions, and address limiting factors, IHH moves closer to its goal of ensuring clean and potable water for all Tanzanians.

Want to Follow Along?

GitHub | LinkedIn | Twitter