Rajeev095/Analysis-of-Broken-and-Non-Functioning-Water-points-in-Tanzania

Constructed classification models to identify working and non-working wells to provide safe drinking water to Tanzanians. # Identifying Broken and Non-Functioning Water-points in Tanzania ## Background Basic access to clean water for drinking and sanitation remains a serious problem for 10% of the world’s population. Tanzania, in sub-Saharan Africa, is number 7 in a recent list of countries without ready access to safe drinking water. In fact, 43% of the countries 57 million people lack such access. Although there has been an exponential increase in the number of water-points (wells, springs, surface water, rainwater collection, etc.) constructed in the past 20 to 30 years, many of them are either no-longer functional, or are in need of repair. Having a predictive model that can classify existing water points as either functional or not, can help the government, NGOs and donors to prioritize and fix or reclaim poorly or non-functioning water-points. ## Source Data The data set was provided by the Tanzanian Ministry of Water and a non-profit group Taarifa. It consists of information on water-points established between 1960 and 2013. The data set contains 59,400 records with 41 variables. The target variable, “status_group” contains 3 classes of values: * 0 = functional (55%) * 1 = non-functional (38%) * 2=functional but needs repair (7%) ## EDA Initial data exploration indicated that many of the variables were different groupings of similar concepts. To get a better handle on this, a mind-map was created. This helped to identify 15 redundant or potentially less useful features. The remaining features were visualized to examine differences in distributions for each of the target classes. For example, it can be seen that older water-points constitute a larger proportion of non-functional ones. Other issues identified and addressed during data exploration and cleaning included: * 7 columns that contained null values were dropped. * Outlier values were examined and in some cases dropped (longitude for example). * About 5 of the continuous features had a large percentage of 0 va …