Machine Learning-Based Quality Prediction and Process Control for the Industrial Production of Palm Oil in Nigeria

Industrial palm oil production involves multiple interconnected unit operations whose performance is influenced by raw material variability and process parameter interactions that are difficult to capture with conventional mechanistic models. Machine learning provides data-driven approaches to quality prediction and process control that may overcome these limitations, yet their application to palm oil processing in Nigeria has not been systematically investigated. This study develops and validates machine learning models for predicting the quality attributes of crude palm oil and for informing real-time process control at a commercial palm oil mill in Rivers State, South South Nigeria. Quality attributes predicted include free fatty acid content, moisture content, colour (Lovibond units), peroxide value, and bleachability index. Input features include fresh fruit bunch ripeness classification by image analysis, sterilisation cycle parameters, digester temperature, press speed, and clarification temperature, collected over 18 months of operational data comprising 4,200 production batches. Five machine learning algorithms, specifically gradient boosting, random forest, support vector regression, artificial neural networks, and long short-term memory recurrent neural networks, were systematically evaluated and compared. The long short-term memory model achieved the lowest root mean square error for free fatty acid prediction at 0.18 percent, capturing temporal patterns in process variability. A reinforcement learning-based process control agent was trained to minimise free fatty acid content in real time and demonstrated a 15 percent reduction in out-of-specification production frequency. The study contributes original machine learning architectures for palm oil quality prediction and represents the first application of reinforcement learning to palm oil process control. Keywords: machine learning, palm oil quality, process control, reinforcement learning, neural networks