Low-cost sensor-based algal bloom labeling: a comparative study of SVM and logic methods

This study explores a low-cost sensor system for real-time algae bloom detection and water management. Harmful algal blooms (HABs) threaten water quality, ecosystems, and public health. Traditional detection methods, like satellite imagery and unmanned aerial vehicle (UAV), are expensive and not always suited for real-time monitoring. The proposed system uses sunlight and illuminance sensors to predict algae blooms and water level fluctuations. Data from these sensors are analyzed using support vector machines (SVM) and logical algorithms, with labeling based on sensor readings (e.g., "algae", "sunny", "shade", "aqua"). A multiple linear regression (MLR) model is also developed to predict chlorophyll-a (Chl-a) concentrations. Both SVM and logical algorithms proved effective. The classification using SVM with four sensor values achieved an accuracy of 92.6%, while applying principal component analysis (PCA) before SVM classification resulted in 91.0% accuracy. In contrast, applying a sequential logical sequence to the boundary conditions of a single SVM model improved accuracy to 95.1%, and incorporating PCA-transformed SVM boundary conditions achieved 100.0% accuracy. This surpassed the performance of nonlinear decision models such as random forest and gradient boosting, which achieved 99.2% accuracy. The MLR model successfully predicted Chl-a levels with a 14.3% error rate for values above 5 mg/L. The developed system is an efficient alternative to traditional methods, enhancing real-time monitoring in water quality management.