Early prediction of reading disability using machine learning.

Area of Science:

• Educational Psychology
• Computational Linguistics
• Machine Learning

Background:

• Early identification of reading disability is crucial for timely intervention.
• Traditional screening methods may lack sufficient predictive power.
• Machine learning offers potential for improved early prediction models.

Purpose of the Study:

• To apply machine learning classifiers for early prediction of reading disability in first graders.
• To evaluate the effectiveness of various feature selection algorithms in identifying key predictive variables.
• To develop interpretable models for clinical and educational use.

Main Methods:

• Utilized a dataset of 356 first graders.
• Applied multiple machine learning classifiers: Support Vector Machines (SVM), Decision Trees (CART, C4.5), Linear Discriminant Analysis, k Nearest Neighbor, and Naïve Bayes.
• Employed Markov Blanket (HITON-PC, HITON-MB) and wrapper-based feature selection algorithms.
• Developed a method to generate interpretable decision tree models from SVM models.

Main Results:

• Support Vector Machines (SVM) achieved an Area Under the Curve (AUC) score greater than 0.9.
• High predictive accuracy was obtained using a limited set of demographic and screening variables.
• Feature selection algorithms successfully identified relevant predictors for classification.

Conclusions:

• Machine learning, particularly SVM, demonstrates high efficacy in the early prediction of reading disability.
• The developed models can aid in identifying at-risk children with minimal data requirements.
• The integration of interpretable decision trees enhances the practical application of these predictive models.