Hybrid statistical and machine-learning approach to hearing-loss identification based on an oversampling technique

Area of Science:

• Genetics and Genomics
• Computational Biology
• Medical Informatics

Background:

• Hearing loss is a significant global health issue impacting spoken language and cognition.
• Early identification of recessive hearing loss features is crucial for preventing disease progression and mitigating social/professional hardships.
• Developing effective decision-support mechanisms is vital for clinical diagnosis.

Purpose of the Study:

• To develop a hybrid statistical and machine-learning approach for predicting hearing loss disorders.
• To create a decision-support mechanism to aid clinicians in diagnosing hearing loss.
• To identify key features associated with recessive hearing loss.

Main Methods:

• A three-phase hybrid approach combining statistical and machine learning techniques.
• Feature selection using stepwise method and random forest (RF) technique.
• Oversampling with synthetic minority oversampling technique (SMOTE) to address data imbalance.
• Classification using logistic regression, multilayer perceptron, and support vector machine (SVM).

Main Results:

• Stepwise method selected three features, RF selected seven features in phase I.
• SMOTE technique significantly improved predictive capability by balancing data.
• Hybrid SVM with stepwise technique demonstrated competitive accuracy, precision, recall, and F1 score.
• SVM models combined with stepwise or RF techniques showed superior Area Under the Curve (AUC) performance.

Conclusions:

• Hybrid approaches combining SVM with stepwise or RF techniques offer higher efficiency than multivariate models.
• The proposed model requires fewer predictors and achieves competitive performance metrics (accuracy, precision, recall, F1, AUC).
• This hybrid model serves as a valuable screening tool for upfront forecasting in clinical practice, capable of identifying vital features and predicting hearing loss.