rahulakrish/water_pump_classifier

Supervised machine learning models to monitor the condition of installed water pumps across Tanzania # Water Pump Classifier ## Description To help the Goverment of Tanzania monitor the condition of installed waterpumps across the country.Given a set of parameters, the model should be able to predict the status of a waterpump. Status can be as classified as: 1. Functional 2. Functional needs repair 3. Non functional ## Dataset Dataset sourced from drivendata.org ## Methodololgy Different classification models were built and the model with the best f1-scores was chosen for optimization. ## Mapping waterpump distribution We can that there are pockets through out the country where there are no pumps installed. This is could be because of geographical features like a desert,mountain range etc. From the map, we can see there seems to be an even spread between the functional and non-functional pumps except for the south eastern region where the pumps seem to be predominantly non-functional. Pumps that are functional but need repair are sparsely represented ## Non-functional Pump deep-dive Before modelling, it is worth looking at non-functional pumps specifically and see if there are any patterns or outstanding factors that lead to their failures: #### Non-functional pumps by Age The older the pump, the more likely it is to fail. We can run a histogram plot to check the rate of failures by age of the pumps: From the plot, it is quite clear that there is a gap in data collection with majority of the pumps having no information about when they were installed. It is also surprising to see that the newer pumps have a higher failure rate compared to the older pumps. #### Non-fuctional pumps by Installer From the data, there are 1201 companies that have installed the pumps.Are some installers have a higher failure rate than others? Visualizing all 1201 companies is difficult, but we can look at the top 20 companies with the most failures: It is quite clear to see that pumps install …