An IOT based Machine Learning Model for Monitoring, Predicting and Forecasting of Demand-side Water Consumption in Nakuru County by comparing Online and Offline Machine learning Algorithms

Water Analytics Model

Abstract

As Nakuru County population continues to soar pressure is mounting on existing water resources to service the ever growing population. The current water resources are already strained, this is characterized by unplanned water rationing -sometimes long spells without the seemingly scarce resource and water borne related outbreaks like the recent case in Umoja area-Nairobi County subsequently impacting on health services, among other issues. The challenge is to systematically ensure provision of water in the increasingly congested urban area of Nakuru County, Landon ward through demand-side water management approach.

Machine learning approaches aim to leverage the volume of data recorded and stored by water service providers to build relevant data analysis, classification, and prediction models for both customer value and business value. Unlike the existing manual, corruptible and labour intensive solution for water distribution and rationing, data-driven machine learning models provide an advanced approach that interrogates both the population and water resource at temporal steps and determines a rationale for water management and commercialization. Hence, this study aims twofold: first, to predict periodic water consumption by machine learning approaches for improved water plans, second, to build a Minimum Viable Product as a real-world solution. More importantly, this will be achieved by comparing ARIMA algorithm and Facebook’s Prophet Algorithm with Incremental–Long Short Term Memory Networks using two different approaches of learning (Online vs.   Offline learning-automating python script vs. manual training) as we seek to identify which combination is most favorable for building and deploying a high performance and highly accurate model for water analytics.

Finally, a comprehensive discussion outlining the main findings, advantages and disadvantages of the ML approaches, future directions of the research and practical insights of this research is reviewed.