toastdeini/Waterpoint-functionality-in-Tanzania

Ternary classification of waterpoint functionality in Tanzania; modeled using data from Tanzanian Ministry of Water. # Predicting Waterpoint Functionality in Tanzania *(Lake Victoria in Tanzania - image courtesy of thepinkbackpack.com.)* ## Overview About 4 million of Tanzania's 59 million people lack access to potable (drinking) water; an even greater proportion of the Tanzanian population (nearly half) lack access to what water.org calls "improved sanitation". While the majority of Tanzanians *do* have access to clean water, a quick look at the functionality status of about 60,000 waterpoints (wells, pumps, etc.) indicates that nearly **half** of those waterpoints either need repair to function without interruption ***or*** don't function at all. ## Problem The Tanzanian Ministries of Water and Health have received a grant from the World Health Organization (WHO) to be used in refurbushing the country's existing water sources. However, the grant cannot sufficiently finance a complete overhaul of the water supply system - time and resources are limited, and the field directors responsible for selecting **where** to position their technicians will need to make judicious, informed decisions about which waterpoints are worth visiting. Of course, the Ministries and their staff/technicians will *not* always know if a waterpoint is functional prior to actually visiting the waterpoint - this is where the machine learning modeling comes in handy. Given a set of descriptive attributes about the waterpoint, but not knowing its true functionality status, I was able to construct a model that accurately predicts a waterpoint's functionality status with 79% accuracy. ## Data Data for this project comes from an ongoing competition hosted by DrivenData, *Pump it Up: Data Mining the Water Table*, which uses data from **Taarifa** and the **Tanzanian Ministry of Water**. Descriptions of each column in the dataset can be found at this link within this repository. The full dataset has forty columns (potential features), each containing information about various attributes of a waterpoi …