Instar identification and weight prediction of Ostrinia furnacalis (Guenée) larvae using machine learning

The Asian corn borer, Ostrinia furnacalis (Guenée), emerges as a significant threat to maize cultivation, inflicting substantial damage upon the crops. Particularly, its larval stage represents a critical point characterised by significant economic consequences on maize yield. To manage the infestation of this pest effectively, timely and precise identification of its larval stages is required. Currently, the absence of techniques capable of addressing this urgent need poses a formidable challenge to agricultural practitioners. To mitigate this issue, the current study aims to establish models conducive to the identification of larval stages. Furthermore, this study aims to devise predictive models for estimating larval weights, thereby enhancing the precision and efficacy of pest management strategies. For this, 9 classification and 11 regression models were established using four feature datasets based on the following features geometry, colour, and texture. Effectiveness of the models was determined by comparing metrics such as accuracy, precision, recall, F1-score, coefficient of determination, root mean squared error, mean absolute error, and mean absolute percentage error. Furthermore, Shapley Additive exPlanations analysis was employed to analyse the importance of features. Our results revealed that for instar identification, the DecisionTreeClassifier model exhibited the best performance with an accuracy of 84%. For larval weight, the SupportVectorRegressor model performed best with R² of 0.9742. Overall, these findings present a novel and accurate approach to identify instar and predict the weight of O. furnacalis larvae, offering valuable insights for the implementation of management strategies against this key pest.