Optimization of CART Models Using Metaheuristics for Predicting Peach Firmness

Abstract

The current advancements in the field of machine learning can have an important application in agriculture and global food security. Machine learning has considerable potential in establishing knowledge-based farming systems. One of the main challenges of data-driven agriculture is to minimize food waste and establish more sustainable farming systems. The prediction of the right harvest time is one of the ways to obtain the mentioned goals. This paper describes multiple machine learning algorithms that are used to predict peach firmness. By accurately predicting peach firmness based on various peach measurement data, a more precise harvest time can be obtained. The evaluation of nature-inspired metaheuristic optimization algorithms in enhancing machine learning model accuracy is the primary objective of this paper. The possibility of improving the peach firmness prediction accuracy of regression tree models using various metaheuristic optimization techniques implemented in GA and metaheuristicOpt R packages is studied. The RMSE on test data of the default regression tree model is 1.722285, while the regression tree model optimized using the gray wolf optimization algorithm scored the lowest RMSE of 1.570924. The obtained results show that it is possible to improve the peach firmness prediction accuracy of the regression tree model by 8.8% using the described method.

Keywords

• machine learning
• nature-inspired metaheuristics
• regression trees
• peach firmness prediction
• food quality