The Development of a Smartphone Application for Water Monitoring in South Africa Using Machine Learning Techniques

Abstract

Water quality monitoring is crucial in South Africa, where water scarcity and pollution present significant public health and environmental challenges. Current monitoring methods are costly, slow, and inaccessible to rural and underserved communities. This study investigates the development of a smartphone application that uses machine learning and computer vision to perform real-time, automated analysis of water quality test strips. The application focuses on six key chemical parameters: pH, total chlorine, free chlorine, total hardness, total alkalinity, and bromine, adhering to South African National Standards. By integrating convolutional neural networks into water quality monitoring, this research addresses the gap in accessible, low-cost solutions. The research employed a supervised learning approach, developing a CNN-based model trained on a dataset of water test strip images collected under diverse environmental conditions. Data pre-processing techniques such as glare removal and brightness adjustments were applied to enhance model robustness. Experimental testing was conducted in ideal conditions (laboratory settings) and non-ideal conditions. Evaluation metrics included accuracy, F1 score, and recall, with thresholds set at 90% for ideal conditions and 80% for non-ideal conditions. User testing assessed the application’s usability and adaptability. The CNN model achieved an accuracy of 92.5% in ideal conditions and 84.7% in non-ideal conditions, demonstrating its effectiveness in diverse scenarios. Key results indicated that pH and chlorine were the most accurately detected parameters, with mean squared errors of less than 0.05. User feedback highlighted the application’s ease of use, with over 85% of participants reporting satisfaction with the app interface and instructions. Findings underscore the potential of AI-driven solutions to revolutionise water quality monitoring in South Africa. The system’s performance exceeds the minimum requirements for practical deployment, though challenges such as lighting conditions and test strip variability remain. The study aligns with global trends in sustainable water management, advocating for community-driven solutions that leverage citizen science. Recommendations include scaling the application to include additional parameters, integrating geotagging features, and enhancing user training modules. This research presents a novel, accessible tool for water quality monitoring, bridging the gap between advanced technology and real-world application in resource-limited settings. The smartphone application offers a scalable model that can empower communities, improve public health outcomes, and contribute to South Africa’s sustainable development goals. Future work should focus on expanding the dataset, improving model generalisability, and exploring partnerships for widespread implementation.