A hybrid approach to predicting and classifying dental impaction: integrating regularized regression and XG boost methods.

Introduction: Dental impaction is a significant clinical challenge that requires advanced predictive modeling and healthcare analytics approaches. Impaction, a tooth alignment issue, is diagnosed using radiographic measurements like panoramic radiographs and CBCT. Artificial Intelligence (AI) is improving the accuracy of predicting dental impaction. Advanced predictive models like logistic Regression and XGBoost analyze critical variables, identify patterns, and perform predictive analysis. These models can identify potential impactions, assess impaction type, and develop treatment plans. Integrating AI into radiographic assessments is expected to enhance further the precision and risk-minimizing capabilities of surgical planning in dentistry. This study presents a hybrid approach combining regularized regression and ensemble methods to enhance the classification and prediction of dental impaction outcomes. By leveraging machine learning and statistical learning techniques, we aim to develop a robust clinical decision support system for dental practitioners. Methods: This research aims to predict the eruption of 3rd molars in the mandible by analyzing three parameters: the distance from the lower 2nd molar to the anterior border, the mesiodistal width of the third molar, and the distance from the apex of the root to the inferior border of the mandible. The study is quantitative, observational, and cross-sectional retrospective. The distance from the lower 2nd molar to the anterior border determines the importance of space available for eruption. The distance from the root apex to the lower border addresses natural eruptive forces and resistance during the eruption. The study aims to find a correlation between eruption and distance from the root apex to the lower border of the mandible. Our feature selection process utilizes ensemble learning algorithms integrated with regularized regression techniques to analyze various parameters. This data analysis framework combines multiple predictive modeling approaches to achieve optimal results. Results: The horizontal type of impaction has the lowest S/W ratio (0.9267), indicating the least available distal to 2nd molar space. This suggests a low potential for future eruptions. The regression equation calculates the S/W ratio using impacted molar width and distal space. A ratio greater than 1.1 indicates a good probability of lower 3rd molar eruption, while a below 0.8 indicates no eruption. The algorithm development process demonstrated the effectiveness of our hybrid approach in dental health analytics. The study improved impaction prediction accuracy to a rate of 78%, with horizontal class predictions achieving a precision of 0.72 and an error rate of 28.1%. Additionally, the regularized logistic regression model attained 75% accuracy for classification and prediction. Conclusion: The study aims to improve dental research by predicting the eruption behavior of lower molars, enabling dental practitioners to make more concise treatment plans. The study identifies the most significant parameters for establishing the space/width ratio: Distance from the second molar to the anterior ramus border and the third molar's mesiodistal width. Enhancing data quality, refining feature selection, and using advanced modeling techniques are crucial for improving predictive capabilities. The findings can help practitioners optimize treatments and reduce potential complications.