The hospitality industry in Yogyakarta faces challenges in maintaining hotel facilities efficiently, prompting the need for data-driven maintenance classification systems. This study evaluates the effectiveness of reducing eleven hotel maintenance indicators using Principal Component Analysis (PCA) and analyzing its impact on Artificial Neural Network (ANN) performance. PCA reduced the indicators to nine principal components while retaining 93% of the data variance, and these components were used to train a multilayer perceptron ANN model. The PCA-based model achieved 90.57% accuracy and a macro-F1 score of 0.9106, slightly lower than the original model using all indicators, which attained 94.34% accuracy with fewer misclassifications. This research provides the first empirical assessment of PCA-based dimensionality reduction in the hotel maintenance context, revealing that although PCA enhances computational efficiency, it can remove subtle yet important information crucial for precise classification. The results highlight that PCA alone may be insufficient for high-precision maintenance prediction and should be complemented with alternative or hybrid feature-selection methods. For hotel managers, the findings emphasize the importance of complete and informative input variables in AI-based maintenance systems, which may be further strengthened through real-time IoT data integration.
Pratama et al. (Tue,) studied this question.