A Statistical Learning-Based Clustering Model With Features Selection to Identify Dyslexia in School-Aged Children.

Area of Science:

• Cognitive Psychology
• Developmental Psychology
• Biostatistics

Background:

• Dyslexia diagnosis is complex, requiring models that capture cognitive skill interplay.
• Traditional statistical methods often oversimplify dyslexia's heterogeneity.
• A robust statistical approach is needed for accurate dyslexia classification.

Purpose of the Study:

• To introduce and evaluate a model-based clustering framework for dyslexia classification.
• To apply finite mixtures of contaminated Gaussian distributions to dyslexia research.
• To identify key cognitive variables associated with dyslexia using variable selection.

Main Methods:

• Employed a model-based clustering framework using finite mixtures of contaminated Gaussian distributions.
• Integrated variable selection techniques to identify clinically relevant cognitive skills.
• Analyzed data from 122 children (51 with dyslexia) in Poland.

Main Results:

• Multivariate finite mixture models demonstrated superior clustering accuracy.
• The model effectively distinguished between dyslexic and control groups.
• Identified Reading, Phonology, and Rapid Automatized Naming as important predictive variables.

Conclusions:

• The multiple-deficit model combined with robust statistical techniques enhances dyslexia diagnosis.
• Finite mixture models offer a powerful tool for understanding dyslexia heterogeneity.
• This approach advances the clinical understanding and classification of dyslexia.