Machine Learning-Based COD Prediction Enhancement in Microalgae Cultivation with POME Using Imputation Techniques

Accurate COD prediction in microalgae-based Palm Oil Mill Effluent (POME) bioremediation systems is essential for effective process optimization and environmental protection. A critical challenge in real-time monitoring is the limitation of IoT sensors in directly measuring COD values, resulting in data availability restricted to periodic laboratory analysis which creates significant delays in process intervention. This study developed an IoT-based monitoring system integrated with imputation methods and machine learning algorithms to predict COD concentration during microalgae cultivation in POME media. To assess the impact of missing data handling on predictive performance, this research evaluated four machine learning algorithms on datasets prepared using four imputation strategies. Results demonstrated that MICE imputation combined with Random Forest achieved the best performance on per-minute data with R² of 0.99, RMSE of 2.97, and MAE of 0.86, significantly outperforming hourly aggregation which achieved R² of 0.95, RMSE of 32.66, and MAE of 12.22. Proper missing data handling at finer temporal granularity substantially enhanced predictive accuracy, enabling reliable real-time COD estimation without laboratory delays. The findings underscore the efficacy of integrating imputation methods with machine learning approaches to link real-time sensor data and laboratory-confirmed COD measurements in POME treatment facilities.