Quantitative review and machine learning application of refractance window drying of tuber slices

Abstract Refractance window drying (RWD) is a preferred drying technique due to its suitability for heat-sensitive products. Although this drying technique appears promising, it is yet largely unexplored. In this study, the authors provide a review of the existing milestones on RWD using a sample of 40 articles from 2000 to date to quantify the state of investigations across multiple studies and establish specific areas needing further attention. Results show that experimental analyses constitute about 53–59 % of the reported cases, followed by a literature review 24–28 %. Furthermore, 17 % of the total study cases was observed across all modelling categories, with machine learning (ML) techniques constituting only about 8 %. Driven by the outcome, this study thus utilized three ML techniques to model the moisture ratio (MR) of 1.5–4.5 mm thick yam slices, operated over the range of 65–95 °C temperature in an RWD chamber. Unlike the routine procedures, the yam thickness versus air temperature effects on moisture ratio were investigated to determine the more significant factor as well as the air velocity effect or its lack thereof on MR. To investigate the validity window for the entire dataset, all data points were considered, with a training-testing ratio of 7:3 used in each case. For scenario one, prediction based on the yam thickness effect showed a greater influence on the MR. The air velocities at 0.5–1.5 m/s had little effect on MR as compared to the case where air velocity was ignored (i.e., the control case in this study). Also, model accuracy for all tested samples has been determined to be better than 93 %. Insight from this study is to guide in the future design of RW dryers for direct measurement of the moisture ratio of harvested root tubers at various conditions.