Machine learning based hearing loss type and tympanogram prediction model

Type of hearing loss and tympanometry are used in hearing measurements. The use of these tests is routine in clinics, but the interpretation of the measurements can lead to subjective results and errors. In this study, machine learning algorithms including CatBoost, Random Forest, and Support Vector Machine were used to predict tympanometry and types of hearing loss. Data obtained from pure tone audiometry, speech audiometry, and tympanometry tests were used for the prediction process. The results were evaluated through the comparison of the performances of the selected algorithms. The study aims to assess the effectiveness and suitability of machine learning approaches in determining types of hearing loss and tympanogram values. In the analysis, the CatBoost algorithm achieved prediction accuracies of 77% and 81% for tympanogram and types of hearing loss, respectively. The RandomForest algorithm achieved prediction accuracies of 75% and 81% for tympanogram and types of hearing loss, respectively. The SVM algorithm achieved prediction accuracies of 72% and 74% for tympanogram and types of hearing loss, respectively. CatBoost yielded better results compared to the other algorithms. After applying the SMOTE method, the score obtained for predicting tympanogram with the RandomForest algorithm is calculated as 94%, while the score obtained for predicting the type of hearing loss is 89%. Increasing the sample size and performing feature selection can enhance the model's success rate.