litzysj/Tanzania-Water-Well

# Tanzania Water Wells ## Overview The Tanzania Ministry of Water asked King and I to look into their dataset and see how we can calculate the functionality of wells in Tanzania. The are many water wells set up across the country. It is important to know how to predict the functionality of those wells in order to make sure that the citizens of Tanzania have water available. ## Business Problem -Stakeholder: Tanzanian Ministry of Water - True business problem: Predict the functionality of wells by reducing false negatives and false positives, the main focus being on false negatives. - Deliverables: Inference or Prediction? -Prediction - Context: False negative: Well functionality prediction says it is actully functional when it is not functional. False positive: Well functionality prediction says it is not functional when it is functional. We would prefer reducing false positives over false negatives ## Evalutation Metric Which metric would make sense to primarily use as we evaluate our models? - Accuracy - balances the two kinds of errors (but is impractical with imbalanced targets) - Precision - helps reduce false positives - Recall - helps reduce false negatives - F1-Score - balances recall & precision (and is beter than accuracy with imbalaced targets) - ROC-AUC - helps focus on better probability outputs (makes sure our predicted probabilities are better) - We will primarily be focusing on recall and precision ## Data Understanding - Terminology - Functional - Not functional - Needs repair ## Conclusions / Recommendations - Focus on wells constructed in 1997 and prior - The wells built during this time period had more non-functional wells than functional wells - Focus on features that better predicted our target variable (non-functioning wells) - Construction year - Source - Region ## Next Steps - Better understanding of data - Improving false positives to improve our prediction # Additional Information * Presentation * Data