Smart Handpumps: A Preliminary Data Analysis

Groundwater accessed by handpumps is the primary water supply for many people in Africa. This “Smart Water” project considers the study of a region of Kenya where there is significant demand for groundwater, especially among the poor. Some of the engineering aims of this project are to determine if data acquired from accelerometers mounted in hand-pumps can be used to perform three tasks: (i) estimate the depth of the groundwater at the pump, (ii) predict pump failure, and (iii) classify the user of the pump (e.g., as being a man, woman, or child). This paper describes an initial investigation, based on one week of data collection, that demonstrates there is useful information in the accelerometer data collected from handpumps, which can be discovered using machine learning techniques. We show that features derived from the accelerometry data exhibit stable, similar behaviour suggesting that users and pump locations may be characterised. We demonstrate that a machine learning system can classify the data according to person and pump and accurately differentiate between different users. We conclude that our preliminary study suggests that information may exist in accelerometry from handpumps that could allow us to answer the three main questions of the “Smart Water” project, described above, motivating a largescale' data-collection activity.