Water potability classification: an approach using machine learning in an embedded system

Abstract

Access to clean drinking water is a vital resource and a recognized human right. However,billionsofpeoplestillsufferfromthelackofaccesstosafedrinkingwater,leading to various health issues. One major challenge in water quality monitoring is the collection and analysis of large a mounts of data. Machine learning models have been widely appliedin water quality monitoring to aid decision making by managers and prevent contamination. An embedded system that integrates sensors with a Machine Learning mode lcould providere al time responses and be feasible for deployment anywhere, regardless of internet connectivity requirements. Such a system, in the context of water potability classification, would allow faster responses to potential threats. This study proposes an energy-efficient TinyML model for classifying water potability, using only parameters available through electronic sensing. The study evaluated performance using metrics such as Accuracy, Precision, Recall, F1Score,memoryoccupiedbythemodel,executiontime,andenergyconsumption,comparing models developed with Random Forest and Neural Networks algorithms. It also assessed the best combination of model and adaptation library for the embedded system. The initial Machine Learning model, using Random Forest, demonstrated good performance, reaching a Precision of 0.70, and compared to its cloud-based counterpart, it can operate for years on a standard battery power source. When comparing all models and libraries in the study, the multilayer perceptron model with the EmbML library used the least memory, with 283,113 bytes, and the Random Forest model with Micromlgen had the lowest energy consumption, usingonly104.534millijoules. This work can help researchers and professionals implement water potability classification systems and use TinyML in other classification problems as well.