Machine learning models and experimental evaluation on mass and temperature of dried coriander seeds in solar hybrid collector with dryer (SHCD)

Abstract

The aim of this study is to examine the drying processes of coriander seeds experimentally and by open sun drying. Using machine learning models, namely Gaussian Process Regression (GPR), Multilayer Perceptron (MLP), and Radial Bias Function (RBF), the experimental outcomes may be assessed. The quality of coriander seeds is greatly impacted by post-harvest drying, yet conventional open sun drying has drawbacks such as contamination and weather dependence. This study uses three machine learning models the Gaussian Process Regression (GPR), Multilayer Perceptron (MLP), and Radial Basis Function (RBF) to predict Heat induced dryness and Mass based dryness in order to experimentally compare sunlight drying (31–58 °C) to open sun drying for 3 kg of coriander seeds. Finding the best model to maximise drying conditions is the aim. The following drying parameters were tracked: air outlet temperature, moisture loss, and sun radiation. R2, RMSE, and MAPE were used to evaluate the model's performance. Important findings were RBF achieved R2 = 0.98 (temperature) and 0.99 (Mass based dryness), outperforming MLP and GPR. Higher mistakes were seen in MLP (R2 = 0.95) and GPR (R2 = 0.56 for mass), which were explained by the sensitivity of GPR to noise and MLP's reliance on the volume of training data. The results are based on 3 kg of drying on a small scale; confirmation on an industrial scale is required. Data on open sun drying, such as humidity and sunlight flux, vary by region. Data noise or inadequate hyperparameter tuning might be the cause of GPR's subpar performance. Results might not apply to seeds with various moisture profiles.