hannah-schurman/tanzania-water-pump-project

Predicting Water Pump Functionality in Tanzania # Tanzania Water Pump Prediction Modeling # Overview --- 4 million people in Tanzania lack access to safe water. Access to clean water is a fundamental human necessity and not having that access is a serious health risk. Our initial dataset shows that almost 40% of water pumps throughout regions of Tanzania are not functional, and an additional 7% need repairs. In this project, my goal was to create a model that would most accurately predict which water pumps are functional, which need repairs, and which need to be replaced completely. To make these predictions, I started with a baseline machine learning model, then tested and tuned more efficiant models and hyperparameters to find the optimal prediction model. The Random Forest model was most efficient and was able to produce a score of 78% taking into consideration the decision to prioritize False Negatives. # Business Problem --- Using data gathered from Taarifa and the Tanzanian Ministry of Water, I was tasked with analyzing the different features corresponding to functional and non functional water pumps with the goal of creating a model that can predict if a pump needs to be replaced. The stakeholder for this project was the World Health Organization and the data was provided by the Taarifa waterpoints dashboard. Through my analysis it is important not to just create an accurate prediction model, but to assess our criteria for what an accurate model will look like. Because we are dealing with human health and life, it is important that we use a conservative model. We want to prioritize replacement over not replacement because our business cost is peoples well being. Therefore, for any false predictions our model makes, we want to prioritize False "not functional" (False Negative) over False "functional" (False Positive). This is because the risks are higher for not replacing a pump that needs replacing, over replacing a pump that didn't actually need to be replaced. This is so we can ensure that clean, …