PERFORMANCE COMPARISON OF SMOTE-BASED MACHINE LEARNING MODELS ON UNBALANCED DATASETS: A STUDY ON DATE AND PISTACHIO FRUITS

Authors : Fatih Bal, Fatih Kayaalp

Pages : 176-200

Doi:10.55071/ticaretfbd.1597150

View : 63 | Download : 45

Publication Date : 2025-06-30

Article Type : Research Paper

Abstract :Creating balanced datasets is a significant challenge that substantially affects the performance of machine learning models in the classification of agricultural products. In this research, we tried to overcome this challenge by using an unbalanced dataset containing information on 7 date palm (Phoenix dactylifera L.) and 2 pistachio (Pistacia vera L.) cultivars. The aim of the study is to compare the classification performance of machine learning models on an unbalanced dataset and a balanced dataset using the SMOTE technique. Initially, classification was performed on the unbalanced dataset using machine learning approaches. Among the machine learning models applied on the unbalanced dataset, the Linear-SVM model showed the highest accuracy rate with an accuracy rate of 92,62%. In the data set extended by applying the SMOTE technique, the RBF-SVM model again showed the highest accuracy rate with 95,55% accuracy rate. In summary, our study highlights the difficulties in machine learning-based agricultural crop classification due to data unbalances. Utilizing the SMOTE technique for oversampling was effective in overcoming this obstacle and improving classification accuracy.
Keywords : Makine öğrenmesi, SMOTE, meyve sınıflandırma, aşırı örnekleme

ORIGINAL ARTICLE URL