Quick Clean Water: IoT and Machine Learning-Based Water Contamination Detection System

Abstract

Quick Clean Water can detect water contamination in private wells, piped water, surface water, and water used for agriculture, recreation, and other purposes in developed and developing countries. Current testing systems are slow, costly, low in availability, and give back minimal results of 10-16 contaminants using expensive strips. Quick Clean Water is reusable, easy-to-use, portable, affordable, and gives advanced results of presently 21 contaminants. First, the Quick Clean Water device calculates the pH, turbidity, temperature, total dissolved solids, conductivity, and salinity using sensors. Inputting the pH, turbidity, conductivity, and TDS, a machine learning model using the Random Ensemble algorithm predicts whether the water is safe at a 55% accuracy rate, which can be improved through data augmentation. Several algorithms were tested and evaluated by the precision, recall, f-score, specificity, negative predictive value, and accuracy rates. The hypothesis was that the K-Means Clustering would result in the best model, but Random Ensemble was the most efficient. If the water is classified as non-potable, users can enter the odor, color, and taste of their water into a 99% accurate ML model using the Random Ensemble algorithm to identify the exact contaminant in their water, which can be advanced by researching more contaminants.