Identifying Children With Clinical Language Disorder: An Application of Machine-Learning Classification

Area of Science:

• Developmental Psychology
• Speech-Language Pathology
• Machine Learning in Healthcare

Background:

• Early identification of language disorders in preschool children is crucial for timely intervention.
• Clinical identification relies on various child and family factors, but optimal classification methods are still being explored.

Purpose of the Study:

• To identify child- and family-level characteristics most strongly associated with the clinical identification of language disorder in preschool-aged children.
• To apply machine learning techniques to classify children receiving language therapy.

Main Methods:

• A sample of 483 preschool-aged children (3-5 years old) was analyzed.
• The least absolute shrinkage and selection operator (LASSO) classification approach was used with background data, including caregiver and teacher ratings.
• The data was split into training (67%) and testing (33%) sets to evaluate classification accuracy.

Main Results:

• The machine learning model achieved excellent classification accuracy (AUC of .87 training, .85 testing).
• Key predictors for language therapy receipt included cognitive impairment, age, gender, and parent/teacher-reported communication, social, and literacy skills.

Conclusions:

• Machine learning provides a robust method for classifying children with and without language disorders based on available background data.
• This approach can aid in identifying critical factors differentiating children who benefit from language services.