Potential Risk Factors Associated with Their Effects Through Classification Algorithms: A Study of Periodontitis on Bangladeshi Garment Sector

Abstract

Abstract Background Periodontitis is a prevalent oral health condition worldwide, posing a significant challenge that requires early detection and intervention to mitigate its adverse effects. This study specifically focuses on understanding the risk factors associated with periodontitis within the Bangladeshi garment sector. By employing ML models, valuable insights can be gained into the variables that contribute significantly to periodontitis risk, leading to the development of targeted preventive strategies and interventions. Methods To achieve the study’s objectives, 12 ML models were selected for analysis, and their variable importance was assessed using 8 of the models. A 2-step CV was utilized, involving a test-train CV with a 75% training dataset, a 25% test dataset, and a repeated 5-fold CV. Random search with a tuning length of 200 was used as a parameter-tuning technique to optimize the performance of the models. For comparative study and best subset model, LR and backward stepwise LR are used with AOR. Results Among the ML models, gcvEarth, demonstrated its efficacy in identifying HRFs for periodontitis and its classification accuracy is 0.9577. From the highest accurate models, the HRFs are age, number of healthy teeth, missing teeth, HTN, gender, sleeping time, and brushing frequency. By leveraging ML techniques, policymakers, healthcare professionals, and stakeholders can make informed decisions and develop targeted preventive strategies with interventions to improve oral health outcomes among individuals in this occupational sector. Conclusions This study’s findings highlight the potential of ML as a powerful tool for identifying HRFs for periodontitis. The integration of ML models, variable importance analysis using OR and AOR, and CV techniques provides a comprehensive framework for understanding and predicting periodontitis in occupational sectors or populations. Furthermore, optimizing the models through parameter tuning with random search enhances the accuracy and performance of the ML models, leading to the development of effective preventive measures and interventions. Ultimately, these advancements contribute to improved oral health outcomes and overall well-being for the Bangladeshi garment sector as well as for the global aspects.