The Impact of Dimensionality Reduction Techniques on Classification Performance in Dry Bean Variety Classification Using Machine Learning

In recent times, the increase in the human population has led to a growing demand for food. In response to this increasing demand, smart farming systems are being developed to enable efficient and conscious agricultural practices. This study focuses on the classification of agricultural product varieties using machine learning, one of the preferred technologies in smart farming systems, with various classification algorithms. Moreover, the effects of different dimensionality reduction techniques on classification success are examined and compared. The dataset used in the study consists of data from seven types of dry beans and includes ${1 3 , 6 1 1}$ data samples. The classification algorithms employed include K-Nearest Neighbor (KNN), Support Vector Machines (SVM), Gaussian Naive Bayes (GaussianNB), Random Forest (RF), Decision Trees (DT), AdaBoost, and Multi-Layer Perceptron (MLP). The dimensionality reduction techniques analyzed are Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Independent Component Analysis (ICA), and Factor Analysis (FA). The comparison results are evaluated based on metrics such as Accuracy, Precision, Recall, and F1Score. According to the accuracy metric, LDA is identified as the most performant dimensionality reduction technique, and it has been observed that dimensionality reduction techniques generally improve or maintain similar performance in the classification models. The impact of dimensionality reduction techniques is most pronounced in experiments with the AdaBoost classification algorithm. In the original dataset, the AdaBoost algorithm achieves a performance of 0.72 according to the F1-Score metric. However, when dimensionality reduction is applied using PCA, LDA, ICA, and FA the AdaBoost algorithm achieves F1-Scores of $0.83,0.91,0.77$, and 0.87, respectively. The use of dimensionality reduction techniques in conjunction with the AdaBoost algorithm results in a significant improvement in classification performance.