Parametric analysis and soft-computing prediction of sweet potatoes (Ipomoea batatas L) starch drying using machine learning techniques

This study is based on parametric selection and prediction of sweet potatoes starch drying using Regression Tree (RT), Support Vector Machine (SVM) and Neuro-Fuzzy (NF) soft-computing techniques. The drying experiments are conducted at Drying Temperature (DT): 40–60 °C, Drying Time (DTi): 0–780 min, Ambient Temperature (AT): 27.2–30 °C and Relative Humidity (RH): 70–80%. Exhaustive search model is used to determine the most and least relevant drying parameters. NF, RT and SVM programming codes are developed in Matlab 9.2 (2017a) with four, three and two-input variable combination NFs (4-1NF, 3-1NF and 2-1NF), RTs (4-1 RT, 3-1 RT and 2-1 RT) and SVMs (4-1SVM, 3-1SVM and 2-1SVM) for the prediction of the starch drying. Exhaustive NF parametric analysis results show that DT-DTi-AT and DT-DTi are the most influential combined variables for three and two variables combinations respectively. DTi and RH are also the most and least influential parameters, respectively. The 3-1NF with neighbourhood radius 0.7 gave the uppermost correlation coefficient (R2) 0.999; and the lowermost root mean square error as well as mean square error 0.0025 and 0.00000625 respectively. The results obtained show that exhaustive search and 3-1NF models are suitable for the prediction of sweet potatoes starch drying.