hanis-z/Tanzania-Water-Well

Predict the condition of a water wells in Tanzania # Predicting water well functionality in Tanzania **Authors:** Hanis Zulmuthi March 2022 ## Business Problem The Ministry of Natural Resources of Tanzania is starting an initiative to improve their responsiveness and preparedness to water well malfunctions. The analysts from Global Water Alliance is building a Machine Learning model that helps predict whether a water well needs repairs or not. With this predictive model, the government can prepare the resources needed to do the repairs, from materials and communicating with the well’s funders and installers to attend to the water wells before it malfunctions. This way we can avoid water disruptions in communities and ensure they always have access to clean water. ## Data Understanding The data is attained through competition on DrivenData. The dataset contains 57,000 records of water wells in Tanzania. There are 40 features in the dataset and our target feature for our predictive model is `status_group`. The original dataset has been pre-processed and cleaned in this notebook and loaded in here for modelling. ## Results ### Modeling Results #### Baseline - Dummy Regressor The first step in doing our analysis here was to create a baseline model using SKLearn's DummyClassifier model. Our predictive models predict 1 if a water well needs maintenance and 0 if it doesn’t need maintenance.Our baseline model has an accuracy of 0.5 which means it predicts well status correctly 50% of the time. #### Best Model - XGBoost Classifier w/ Hypertuning Our 5th model which is a XGBoost Classifier that is hypertuned is by far our best model with an accuracy score of 89% and a higher AUC score of 0.96. ### Modeling Performance Our best model has an accuracy of 89% and this heatmap shows the average difference between the actual well status vs predicted well status by region. White colors are regions we don't have data for. Darker colors represent no to minimal difference and lighter colors show bigger dif …