<![CDATA[Purpose – This study aims to optimize Support Vector Machine (SVM) parameters using Particle Swarm Optimization (PSO) for sensor-based water quality classification and to integrate the optimized model into a real-time monitoring dashboard. The performance of SVM strongly depends on hyperparameter selection, and improper tuning reduces classification accuracy and generalization. This research argues that PSO-based optimization significantly improves classification performance while supporting operational environmental monitoring. Design/methods/approach – An experimental design was employed using 2,400 sensor observations collected from three river locations over 30 days, measuring pH, turbidity, dissolved oxygen, temperature, and total dissolved solids. After preprocessing, 2,320 valid records were analyzed. Standard SVM (RBF kernel) was compared with PSO-SVM using 5-fold cross-validation. PSO used 30 particles and 50 iterations to optimize C and γ within predefined search ranges. Performance metrics included accuracy, precision, recall, F1-score, confusion matrix, and paired-samples t-test (α = 0.05). Findings – PSO-SVM achieved 93.12% accuracy compared to 83.76% for standard SVM, reducing error rate from 16.24% to 6.88%. The mean accuracy improvement of 9.36% was statistically significant (t = 8.74, p = 0.001; Cohen’s d = 1.82). Optimal parameters were C = 47.83 and γ = 0.023. The dashboard demonstrated 0.12-second classification latency and 99.2% uptime. Research implications/limitations – Data were collected from only three locations and five sensor parameters, limiting generalizability. Originality/value – This study integrates PSO-optimized SVM with IoT-based dashboard monitoring, offering a replicable framework for intelligent and real-time environmental monitoring systems]]>
