APPLICATION OF MACHINE LEARNING IN FRUIT RIPENESS PREDICTION

Abstract

This thesis presents an application of machine learning, specifically deep learning, to predict the ripeness level of fruits using image classification. The study focuses on three common fruits: banana, apple, and orange, classified into three ripeness stages: unripe, ripe/fresh, and rotten. The Fruit Ripeness dataset from Kaggle was utilized, containing approximately 3,600 images divided into nine classes. A convolutional neural network based on ResNet18 with transfer learning was implemented using PyTorch, achieving an overall test accuracy of 95%. The model performed exceptionally well on all classes, with F1- scores averaging 0.96. A practical prediction system was developed, allowing users to upload real-world fruit images for ripeness assessment. The results demonstrate the potential of automated fruit ripeness detection in reducing agricultural waste and supporting harvest decisions. Future improvements include dataset expansion and integration into mobile applications. Keywords: machine learning, deep learning, fruit ripeness prediction, image classification.