Application of Data Mining Algorithms for the Detection of Non-functional Bore Wells

Abstract

India is the largest user of groundwater, with approximately 27.5 million wells producing over 230 cubic kiloliters annually. Artesian wells are vertical that are excavated in the subterranean aquifer on the surface to collect water for a variety of purposes. The number of artesian wells dug in a year has increased due to the reasons such as reduced rainfall, water shortages and depletion of groundwater etc. As the water dries, the motor is removed and the exterior is not properly covered or sealed. The diameter of the borewell is large enough for a child to fall into the interior of the currently inoperable or unused well.  According to one report, more than 40 children have fallen into wells since 2009, and sadly 70 percent of rescue operations have failed. The current paper contains an analysis of the dataset from the drill holes obtained by Kaggle. The objective of this work is to build Predictive models using machine learning algorithms such as naive bays, ANN models, and logistic regression to "predict existing out-of-service wells that will be reported and seized in the near future." i.e. to predict the non-functional bore wells that require immediate actions .The performance of the model was evaluated using performance metrics. Logistic regression shows good performance in predicting killer non-working wells.; K-NN