Modeling and optimization of energy efficiency and product quality in staple food roasting using machine learning: A case study on cassava processing

• Using the Machine Learning Integrated I-Optimal Response Surface Methodology (RSM) Design, lab-scale roasting process data was modelled, optimized and used to predict the design parameters of a real-life solar PV roaster. • The developed Lab-scale and real-life (Solar PV) roasting systems have the potential to reduce specific energy consumption by 80% at standard quality parameters. • By using the Rate of Rise ( RoR ), the roasting process can be achieved in real time, with consistent quality attributes and accurate forecasting of important roasting milestones. • Temperature at 130 °C, 40 RPM agitation speed, 22.27 min and 1.37 kWh kg-gari −1 specific energy consumption, are the optimal at standard quality parameters of cassava roasting to gari. • The developed Random Forest regressor can be trained to design and develop efficient, high-quality roasting systems. Energy-efficient food processing techniques, including roasting could help reduce post-harvest losses which currently exceed 30 % annually. Roasting enhances food dehydrated for extended shelf life and quality. In most developing countries, roasting is drudgery, nutrient altering, unsustainable and mostly woodfuel powered. Lack of optimized important process factors such as temperature and agitation speed, presents a significant challenge in achieving process and energy efficiency, Therefore this research sought to bridge these gaps by modeling and optimizing the temperature/Rate of Rise (RoR) and agitation speed at standard quality of the roast using fresh cassava processing into cassava-pulp as a case study. The Machine Learning (ML)-Integrated Design; I-Optimal Response Surface Methodology (RSM) and Random Forest (RF-Python) techniques with two factors (temperature (90–250 °C and agitation speed-10–50 RPM) and seven responses, were used for experimental design, modeling, optimization, prediction of design parameters for real-life application in a solar PV roaster and model validation. External validation was reinforced with python driven Mann-Whitney U, Levene and Bland-Altman tests. The optimum parameters where find to be; 130 °C-roasting temperature with 21.08 °C min −1 -RoR, 40 RPM, 22.27 min, 1.37 kWh kg-gari −1 -Specific-Energy-Consumption (SEC), color-(74.55- L* , 2.97- a*, −23.98- b* ), 3.40-swelling index-(SI), and 1.36 mm-texture. The models predicted the real-life Solar-PV roasting system design parameters with accuracy above 90 %, Mean-Absolute-Error (MAE-0.99) and showed significant SEC savings of 81.40 % against the 7.37 kWh kg-gari −1 woodfuel-based-cookstove. This scalable, data-driven solution could be applied to design and develop more efficient and high-quality roasting systems. Future studies should focus on generalizing with different staple foods.