Yield prediction of Nsukka yellow pepper using some machine learning models

In this study, the accuracy and efficiency of four machine learning models, Random Forest, Decision Tree, Multiple Linear Regression and Support Vector Machine were evaluated for predicting the yield of Nsukka yellow pepper. This kind of predictive study will help farmers and other related workers in informed decision making with farm inputs, resource optimization, risk management and market planning. The study was carried out at the Research Farm, Department of Agricultural and Bioresources Engineering, University of Nigeria, Nsukka, Enugu State, Nigeria. Historical data for temperature, humidity and solar radiation were collected from the Meteorological Station of the National Centre for Energy Research and Development, University of Nigeria, Nsukka. The historical data covered a period of three months, 9th January to 9th April 2024, corresponding to the period when crop parameters such as plant height, leaf length, leaf width and crop yield were measured and collated for one hundred and seven plants of Nsukka yellow pepper. This period covers the time from transplanting to harvesting. The machine learning regression algorithms were trained, validated and tested using yield data from the pepper farm. The algorithms developed with Python codes were trained in Google Colab. Results revealed that the Multiple Linear Regression model had the best performance metrics with an MSE of 45.23, RMSE of 6.73 and R2 of 0.67. This implies that the Multiple Linear Regression model is an effective tool for predicting the yield of Nsukka yellow pepper, and hence useful for decision support to farmers. Furthermore, to statistically confirm the performance differences among the models, one-way ANOVA and ANCOVA tests were conducted. ANOVA results (F = 25.76, p < 0.001) showed significant variation among the mean predicted yields for the four models, while the ANCOVA (F = 18.92, p < 0.001), controlling for actual yield, confirmed that the models differed significantly even after accounting for real yield effects. This confirms that the Multiple Linear Regression and Support Vector Regression models outperformed Random Forest and Decision Tree models in terms of predictive consistency