Supervised Learning Models for Tomato Quality Prediction Using Electronic Nose Data

The goal of this study is to evaluate the feasibility of using an electronic nose for the quality-based classification of two varieties of tomatoes, Tounvi and Akikon, mostly cultivated in Benin. To achieve this, a device consisting of 8 semiconductor metal oxide gas sensors was used to collect olfactory data on five fruits of each tomato variety for 70 days, until each fruit was unfit for consumption. The collected data was then utilized to train Random Forest, k-NN, SVM and MLP for classification. The obtained results showed that, generally, performances obtained were better on an oversampled dataset. The random forest algorithm exhibited the highest performance, achieving accuracies of 0.92, 0.94, and 0.98 on undersampled, normal, and oversampled data, respectively. Findings indicate that utilizing an electronic nose in combination with a supervised learning model can be a successful approach for classifying different post-harvested tomato varieties based on quality. This method offers a valuable solution that can expedite the classification process and enhance decision-making in the tomato industry.