HughJackmannix/Classifying-Tanzania-Well-Functionality

# Classifying Tanzania Wells Based on Functionality **Author:** Jack Mannix ## Summary A model trained to predict the functional status (‘Functional’ or ‘Non-functional’) of a well in Tanzania allows responsible entities to be notified when a well is in need of repair ## Business Problem WaterTanzania, a nonprofit, wants to improve access to water across Tanzania. To do so, they want to be able to send alerts to the entities in charge of a well(s) to check the status its functional status. Ontop of sending alerts when a well is deemed 'non-functional,' WaterTanzania also wants to send precautionary alerts based on observed well characteristics in order to repair wells before they become 'Non-functional'. With this goal in mind, the nonprofit needs an algorithm that can accurately predict the status of a well, and prioritize minimal false predictions for wells that are actually 'non-functioning'. ## Data and Methods All models were trained and tested on data collected from 47,304 wells - Data provided by TAARIFA and DrivenData - Data collected 2011 - 2013 - Models used 1. Logistic Regression 2. DecisionTree 3. RandomForest 5. XGBoost ### Assumptions The original data provided three target variables 1. Functional 2. Non-functional 3. Funcional-Needs Repair Due to a major class imbalance against 'Functional-Needs Repair', preliminary models produced severly low recall for this variable To address this imbalance, 'Non-functional' and 'Functional-Needs Repair' were combined due to the following considerations: 1. A well that is 'Functional-Needs Repair' is not fully functional, and will decline to 'Nonfunctional' if not addressed 2. Additional issues of misclassification arise: - Wells that are 'Non-functinonal' that are classified as 'Functional-Needs Repair' would be less of a priority to fix - Wells that are 'Functional-Needs Repair' that are classified as 'Functional' miss the chance for being repaired before becoming 'Non-functional' 3. Additional m …