Dental implants success prediction by classifier ensemble on imbalanced data

• Main Authors: Mostafa Sabzekar, Motahare Namakin, Hanie Alipoor Shahr Babaki, Arash Deldari, Vahide Babaiyan
• Format: Article
• Language: English
• Published: Elsevier 2021-01-01
• Series: Computer Methods and Programs in Biomedicine Update
• Subjects: Imbalanced data; Dental implants; SMOTE algorithm; Classifier ensemble; Genetic algorithm
• Online Access: http://www.sciencedirect.com/science/article/pii/S2666990021000203

Background: In recent years, increasing advances in data storage have led to an increase in the volume of medical data. Data mining has a significant role in different aspects of medical sciences, including extracting knowledge from the volume of data, diagnosing diseases, determining the effectiveness of drugs, identifying treatment methods, and predicting the success of treatment. The dental implant is one of the most common methods for tooth root replacement used in prosthetic dentistry. Objective: Accurate prediction of implant success or failure before the surgery is an interesting and challenging topic for dentists. However, imbalanced data is one of the main problems in any medical data mining application. This problem occurs when samples of one or more classes of data are scarce or difficult to collect. This challenge poses severe problems in extracting knowledge from data. Methods: In this paper, we present a new approach for predicting the success or failure of dental implants in the presence of imbalanced data. To this aim, first, the imbalanced data is clustered, and then each cluster is balanced by SMOTE algorithm. In the next step, the balanced data are classified by an ensemble of four well-known classifiers, including the decision tree, the support vector machine, the k-nearest neighbor, and the Naïve Bayes. To improve the classification accuracy, an optimal weight is determined for each classifier using the genetic algorithm. Thus, each classifier will have a different effect on determining the final decision. Results: : The proposed method was applied to a dataset consists of 224 patients who underwent implant and bone graft treatments at the Faculty of Dentistry of the University of Tehran, Iran. Different criteria were used to evaluate the proposed method, including accuracy, sensitivity, specificity, AUC, and G-mean. The evaluation results show that the proposed ensemble algorithm provides a more accurate implant success prediction on imbalanced data compared to using the single-use of each classifier and also state-of-the-art researches. Conclusions: This paper presents a novel ensemble method algorithm to predict the success or failure of dental implants for imbalanced data. The obtained results reported that the use of the hybrid algorithm increases the accuracy parameter by about 5.5%, the sensitivity parameter by about 0.3%, and the specificity parameter with a significant amount of about 25%, in comparison to the best results achieved by single classifiers.