Date Fruit classification using a wide range of classifiers

Date fruits (Phoenix dactylifera) are fruits known for their high nutritional value, and also for their pleasant taste. They have been cultivated in the Middle East and the Indus Valley for thousands of years. There are different types of date fruits, and their classification is performed based on external characteristics such as color, size, diameter and shape. Traditionally, the classification of these fruits used to be done manually. However, the development of automatic classifiers for this task has been attracting attention in recent years. In the automatic classification scenario, machine learning and pattern recognition techniques are used for the development of classifiers. Despite the number of studies that address the classification of date fruits, there is an apparent lack of comparative studies that present evaluations involving different types of base classifiers. In this context, our study considers the classification of date fruits from a comparative perspective, with five learning algorithms: Support Vector Machines (SVM), Multi-Layer Perceptron (MLP), Naive Bayes (NB), K-Nearest Neighbor (KNN), and Logistic Model Trees (LMT). The algorithms were evaluated on three datasets. Two of these sets were derived from the original one by applying data balancing techniques. In general, it was observed a variation in the F-Measure of from 0.832 to 0.961. As far as we know, in this study we obtained the state of the art performance on the “Date fruit dataset” (0.961 of F-Measure using stacking).