Perbandingan Berbagai Metode Segmentasi dan Mechine Learning pada Makanan Tradisional Sumatera Utara

Abstrak

This study investigates the categorization of traditional North Sumatran dishes using various segmentation methods. The goal is to participate and participate in the preservation of North Sumatran culture. The study covers 34 types of traditional North Sumatran dishes originating from various regions. Food images are processed using segmentation techniques such as Sobel, Prewitt, Robert, Scharr, and Canny filters. The data set is then used in traditional machine learning algorithms, including Random Fortst, Decision Tree, and four SVM algorithms, for classification purposes. Among the algorithms with the highest performance, the Random Forest algorithm with Robert's segmentation method achieves outstanding results on dataset testing, with 85.52% accuracy, 84.63% recall, 83.77% precision, and 82.49% f1 score. The execution time for most of the best performing algorithms is around 1 minute on average. In addition, the Random Forest algorithm with the Canny operator achieves 81.51% accuracy, 84.97% recall, 86.81% precision, and 85.61% f1 score on dataset testing. The Random Forest algorithm with the Sobel operator obtains an accuracy of 78.41%, a recall of 65.28%, a precision of 62.33%, and an f1 score of 63.71%. Among the four SVM algorithms, the Sigmoid SVM with the Scharr operator achieves the highest performance in its category across all classification metrics. The importance of insight into the traditional cuisine of North Sumatra is invaluable. Emphasizing the importance of this research in promoting the preservation and introduction of traditional North Sumatran food.