Optimizing Durian Chip Quality Using Machine Learning: Multiple Linear Regression for Predicting Inputs in Microwave-Hot Air Drying Process.

The goal of this study was to develop a model that could predict and optimize inputs for a microwave-hot air drying process for crisp durian chips. Machine learning (multiple linear regression) was used to examine the relationship between the input variables (drying time, microwave power, the initial thickness of durian slices, total solid content) and the outcome variables (hardness, number of peaks, colors). Train-test split and K-fold cross-validation ensured the accuracy of the model, with R-squared values ranging from 0.803-0.976 from K = 5 to K = 10. The performance of the model was evaluated by comparing predicted values to experimentally observed values. The model demonstrated that not only food quality consistency was achieved, but also time-consuming trial-and-error methods were reduced by a remarkable 96 percent. Utilizing these optimal inputs for the model significantly decreased energy consumption across multiple parameters. The blower's energy consumption decreased by 18.2 percent, heat usage decreased by 19.3 percent, and microwave energy consumption decreased by 22.1 percent. Therefore, this model could inspire various food processing methods to increase productivity and food quality.