Forecasting Apple Fruit Color Intensity with Machine Learning Methods

Abstract

In this study, we focused on the possibility of forecasting the development of skin color in apples on the basis of weather forecast by using a machine learning methods. We used supervised learning and generated models via the use of six decision trees. The purpose of the research was to build models that would allow for in-practice-acceptable accuracy in the prediction of the development of fruit skin color (especial a colour parameter a*), for three apple varieties. For cv. ‘Gala, Brookfield’, the most accurate models were generated by using decision tree J48 (89.13% accuracy). For late ripening cv. ‘Fuji, Kiku 8’ and cv. ‘Braeburn, Maririred’, the most accurate model was obtained by using decision tree LMT (91.73 and 96.65% accuracy). The data confirm that the applicability of predictive models strongly depends on the accuracy of weather forecasts. In regard to the seven-day weather forecast, which was used for expert models, the accuracy of the models was, on average, reduced by 10.73%.