Machine Learning based Classification of Contaminated Drinking Water Using Raspberry Pi Embedded System and IoT Device

Abstract

The drinking and human usage fresh water are contaminated and polluted by wastages released from dyeing and other industries. The classification of highly contaminated, mildly contaminated, lightly contaminated and recycled water is important for different level of water purification. The aim of this project is to develop an Internet of Things based low cost embedded Raspberry pi model for drinking water image classification. For image classification, a machine learning based algorithm proposed here called as KNNMPCAF3 (K-Nearest Neighbour Machine Learning based Principle Component Analysis First Three). Initially, the collected samples are arranged in four classes, no contamination, lightly contaminated, moderately contaminated and severely contaminated. The proposed algorithm first collects the top 3 principle components from each collected image samples. Then the principle components of each samples encoded using machine learning algorithm provided by python library scikit-learn. Then they are classified using K-nearest neighbour machine learning algorithm. This model works well for classification of contaminated drinking water and the image samples collected from different regions are trained to give better accuracy of classification from 70% to 100%. If the water is contaminated more severely in a particular area and a siren is alarmed. Then siren immediately followed by a notification message which is conveyed to the municipality or corporation or town Panchayat or village Panchayat by a Panchayat sanitary worker or health inspector.