Fuzy Yustika Manik, Syahril Efendi, Dewi Sartika Ginting, T.H.F. Harumy, S. Kana Sahputra, Nurhayati
Many features are implemented in computer vision and machine learning applications. Because no one knows how a feature affects the classification results of machine learning methods. Therefore, there is a tendency to increase the number of features. However, features may need to be more relevant than some extracted components. In addition, high feature dimensions can reduce performance in classification. In classifying pineapple maturity levels, 12 characteristics were obtained from extracting color, texture, and color characteristics. Utilizing dimensionality reduction, original data information can be preserved while reducing the number of variables. Feature dimension reduction is done using the feature reduction method and the PCA feature selection method. These important dimensional variables are condensed into fewer without sacrificing information in previous data. The five dominant components have been able to cover 94%• So, for this research, five main components were used, which will then be used in the SVM, KNN, and ANN machine learning models. There is an increase in the percentage of accuracy for each machine-learning method used. The increase in accuracy in the SVM and ANN methods is not very significant, only 1% and 2%. However, an increase in accuracy can be seen in the KNN method of 5%. With an increase in accuracy percentage, the PCA method can produce the best features for understanding the data, reducing processing requirements, and improving prediction performance. © 2023 IEEE.
Universitas Sumatera Utara, Faculty of Computer Science and Information Technology, Sumatera Utara, Medan, Indonesia; Universitas Negeri Medan, Faculty of Mathematics and Natural Sciences, Sumatera Utara, Medan, Indonesia; Universitas Mikroskil, Faculty of Informatika, Sumatera Utara, Medan, Indonesia
