Machine learning algorithms to estimate drying characteristics of apples slices dried with different methods

In present study, Golden Delicious, Oregon Spur and Granny Smith apples were dried with the use of five different drying methods (open-sun, controlled greenhouse, microwave oven, hybrid dryer (microwave+convective) and convective dryer) and moisture ratio, moisture content and drying rate values were determined. Then, different machine learning algorithms (artificial neural network, k-nearest neighbors, random forest, gaussian processes and support vector regression) were used to estimate moisture ratio, moisture content and drying rates from drying methods, initial single product thickness, initial single product weight, single product moisture content, drying duration and single product temperatures of dried apples. Present findings revealed that output parameters were accurately estimated by machine learning algorithms. For moisture ratio estimation of all apple cultivars, Random Forest was identified as most successful algorithm with correlation coefficients (R) of 0.9800, 0.9873 and 0.9841, respectively. Random forest also yielded greatest correlation coefficients for moisture content and drying rate estimations.