EAT: explainable attentive transformers for identifying the factors influencing dental visits to enhance dental data completeness

Abstract

Background: Access to routine dental care is a cornerstone of preventive healthcare. Regular dental check-ups, which include professional cleanings, examinations, and preventive treatments, play a crucial role in preventing advanced dental diseases such as cavities, gum disease, and oral cancer. These check-ups help identify potential problems early, reducing the need for more invasive treatments and minimizing complications. This research aims to identify key determinants influencing patient behavior regarding dental care. Methods: To identify influential factors affecting annual dental visits (ADV), we utilized the publicly available 2022 Behavioral Risk Factor Surveillance System (BRFSS) dataset, comprising survey records. This dataset captures health-related behaviors, chronic conditions, and access to preventive services among adults in the United States. We propose a hybrid method combining feature selection using the transformer with machine learning (ML) models to uncover the determinants of ADV behavior. Results: The proposed model was evaluated using various transformer architectures. Among them, RoBERTa, ELECTRA, and BERT demonstrated the highest performance. Features selected by these top-performing models were subsequently used to train several ML models. CatBoost and XGBoost achieved the highest accuracies at 76.0% and 75.6%, respectively, while the decision tree achieved the lowest accuracy at 64.8%. Conclusions: Our proposed method effectively reduced the feature space, thereby improving focus and reducing training and inference time without compromising accuracy. This fusion-based model provides valuable insights for healthcare providers, enabling the development of targeted interventions tailored to specific population needs. Understanding the factors contributing to irregular dental visits can guide evidence-based strategies to overcome barriers and improve overall oral health outcomes.