Exploratory Data Mining Techniques (Decision Tree Models) for Examining the Impact of Internet-Based Cognitive...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Psychological Interventions

Background:

• Tinnitus is a heterogeneous condition with variable treatment responses.
• Current management recommendations often use a "one size fits all" approach.
• Predictive models for tinnitus treatment success are lacking.

Purpose of the Study:

• To identify variables associated with treatment success in internet-based cognitive behavioral therapy (ICBT) for tinnitus.
• To explore the utility of decision tree models for predicting ICBT outcomes in tinnitus patients.

Main Methods:

• Utilized exploratory data mining, specifically decision tree models, on data from 228 individuals across 3 clinical trials.
• Assessed treatment success via a 13-point reduction in Tinnitus Functional Index (TFI).
• Employed Classification and Regression Tree (CART), C5.0, Gradient Boosting (GB), XGBoost, AdaBoost, and Random Forest models, with Shapley Additive Explanations (SHAP) for predictor importance.

Main Results:

• CART and Gradient Boosting models demonstrated the best predictive performance (accuracy ~70-72%, AUC ~0.68-0.69).
• Higher education level emerged as the most significant predictor of ICBT success.
• The CART model identified three distinct participant groups with an 85% probability of ICBT success.

Conclusions:

• Decision tree models, particularly CART and Gradient Boosting, show promise for predicting ICBT outcomes in tinnitus management.
• Future research should incorporate larger sample sizes and a broader range of predictive factors to enhance model accuracy.