Predicting the Cochlear Dead Regions Using a Machine Learning-Based Approach with Oversampling Techniques.

Area of Science:

• Audiology
• Machine Learning
• Data Science

Background:

• Cochlear dead regions (DRs) are crucial to identify in clinical audiology.
• Machine learning (ML) offers potential for predicting DRs.
• Dataset imbalance is a challenge in developing predictive models.

Purpose of the Study:

• To develop and evaluate an ML-based model for predicting cochlear dead regions (DRs).
• To investigate the impact of oversampling techniques, specifically the synthetic minority oversampling technique (SMOTE), on model performance for audiological datasets.

Main Methods:

• Utilized classification tree (CT) and logistic regression (LR) as prediction models.
• Applied the synthetic minority oversampling technique (SMOTE) to address data imbalance.
• Performed 10-fold cross-validation to assess model accuracy.

Main Results:

• Classification tree (CT) achieved higher accuracy (0.93 ±0.01) than logistic regression (LR) (0.82 ±0.02) on original data.
• Oversampling with SMOTE resulted in decreased accuracy for both LR (0.66 ±0.02) and CT (0.86 ±0.01).
• The SMOTE method did not enhance the predictive performance of the ML models.

Conclusions:

• This research is the first to apply SMOTE to audiological datasets for DR prediction.
• Oversampling using SMOTE did not improve the accuracy of the developed ML models.
• Future work should explore more sophisticated models or incorporate additional clinical features for better DR prediction.