Introduction of Songket Motifs from Four Regions Using Inception V3 Image Embedding and Machine Learning in Orange Data Mining

Background: Manual identification of songket motifs is time-consuming and requires specialized expertise, limiting the cataloging and documentation of regional textiles. Objective: To develop a songket motif classification system using image embeddings extracted with Inception v3 and machine learning algorithms implemented in Orange Data Mining. Methods: The research dataset comprised 200 images of songket fabrics representing four regional motif categories: Balinese songket, Lombok songket, Palembang songket, and Riau songket, with 50 images per category. Image embeddings were extracted from all images using Inception v3, and the resulting embeddings were used as inputs for several classification models. The classification algorithms evaluated in this study were Logistic Regression, Support Vector Machine (SVM), Random Forest, and Neural Network. Results: Logistic Regression achieved the best overall classification performance, with an accuracy of 75%, an area under the receiver operating characteristic curve (AUC) of 0.905, and a Matthews correlation coefficient (MCC) of 0.667. Confusion matrix analysis further revealed that Palembang and Riau songket motifs had per-class recall values of 88% and 84%, respectively, whereas Balinese and Lombok songket motifs had the lowest per-class recall values, at 62% and 66%, respectively. High visual similarity between Balinese and Lombok motifs may explain this difference in classification performance. Conclusion: Findings demonstrate potential of combining Inception v3-based image embeddings with machine learning algorithms for classifying traditional textile motifs. They also highlight the potential of implementing this approach in Orange Data Mining to support the digitization and cataloging of traditional textiles, thereby contributing to the preservation of Indonesia's textile cultural heritage.