Modelling Black Soldier Fly Larvae Growth Performance Based on Chemical Composition and Rearing Conditions—A Machine Learning Approach

Abstract Purpose Food waste arising from households, restaurants, business places and other tertiary food waste comprises a diverse range of organic fractions rich in different nutrients (e.g. protein, lipids) suitable for the rearing of black soldier fly larvae (BSFL). The objective of this study was to predict biological age and weight of BSFL reared on different waste streams using machine learning (ML). Methods Samples were collected from different waste streams including bread plus vegetables, soy waste, as well as a combination of commercial waste streams. Results For the prediction of BSFL biological age, the coefficient of determination (R 2 cv ) and the standard error of cross validation (SECV) were 0.76 (SECV: 3.89 days), 0.76 (SECV: 4.27 days) and 0.77 (SECV: 2.76 days) using all samples, bread plus vegetables (BV) and soy waste, respectively. For the prediction of the BSFL weight, the R 2 cv and SECV were 0.56 (SECV: 0.46 g), 0.86 (SECV: 0.21 g) and 0.75 (SECV: 0.38 g) using all samples, BV and soy waste, respectively. Conclusion This study demonstrates how ML techniques, specifically partial least squares regression, can be utilised to predict the biological age and weight of BSFL using a combination of biological and non-biological variables. The utilisation of different types of diets indicates how the combination of different variables plays unique roles in influencing the growth performance parameters of the larvae. The utilisation of chemometrics methods provides tools that allow a better understanding of the nutrition of the BSFL, as well as contributes to improving the efficiency of the production. Graphical Abstract