Eggplant Yield Prediction Utilizing 130 Locally Collected Genotypes and Machine Learning Model

Accurate crop yield prediction has long been a challenge for the agricultural community, with significant consequences for food security, farmer livelihoods, and strategic planning. Traditional forecasting methods based on agro-meteorological data have proven inadequate for capturing the wide variety of factors impacting production. Eggplant is also well-known throughout the world as one of the most popular fruit vegetables, and demand for this vegetable is growing as the world’s population grows. Yield, i.e agricultural production per plant is significantly related to fruit diameter, fruits per plant, percent of fruit infestation by brinjal shoot and fruit borer, and fruit weight attributes, indicating that direct selection based on fruit number and fruit weight may be adequate for improving other traits. This study demonstrates a machine learning approach for precise eggplant yield prediction using data from different regions of Bangladesh, analyzing three regression algorithms such as categorical boosting regression (CatBR), light gradient boosting regression (LGBMR), and extreme gradient boosting regression (XGBR), as well as the Bayesian approach used for choosing optimal hyperparameters. CatBR emerged as the best-performing algorithm for eggplant yield prediction, utilizing 10 genetic parameters. The CatBR model we used performed admirably, displaying both reliability and speed in forecasting crop yields based on the most important variables. It achieved an R
²
score of 68.02%, as well as an MSE score of 3.8117, an MAE score of 1.3090, and a MedAE score of 0.9949. Notably, our proposed model is capable of flawlessly combining and analyzing data sourced from various regions across Bangladesh, ensuring high precision and avoiding model overfitting.