Statistical Learning for Islamic Heritage Identification: Classifying Sumatran Sultanate Affiliation Through Gravestone Characteristics
Year: 2026
Type: article
Abstract: Classifying sultanate affiliation based on gravestone characteristics is necessary to support more objective artifact identification. Similarities in gravestone shape and decoration across regions often complicate affiliation determination through visual observation alone. This study compares the performance of Linear Discriminant Analysis (LDA) and Multinomial Logistic Regression (MLR) in classifying sultanate affiliation based on the physical characteristics and decoration of gravestones in Sumatra. This study used 319 gravestone data from Islamic sultanate burial complexes in Sumatra. Sultanate affiliations were grouped as Aceh, Pagaruyung, Siak, Deli, and Others. The predictor variables included type, shape, material, physical condition, upper, middle, and lower decorative elements, and calligraphy text. All predictors were dummy-coded, and both models were evaluated using repeated stratified 5-fold cross-validation, accompanied by an assessment of LDA assumptions and a sensitivity analysis excluding gravestone type. The cross-validated accuracy was 95.28% for LDA and 93.51% for MLR. The difference between the two methods was not statistically significant, and LDA remained stable despite violating its assumptions. Gravestone type and material were the most influential attributes. Without gravestone type, the accuracy remained at approximately 87%, with material as the strongest marker of affiliation.
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225305115