A machine learning approach for managing the potential risk of odds of developmental stuttering

This study attempts to propose a predictive analytics approach for the prediction of the potential odds of developmental stuttering in preschool-aged children. This approach would yield better and more accurate results for the whole population of children. The publicly available dataset from Vanderbilt University was chosen to acquire research-based historical temperamental data on both children who stutter (CWS) and children who not stutter (CWNS). Machine learning (ML) algorithm, called, Logistic Regression has been implemented for the prediction purpose. The outcomes of this study suggest that the temperamental parameters of CWS and CWNS can be utilized to predict the potential risk of odds of developmental stuttering. The role of feature selection algorithms was found to be very decisive for generating optimal prediction results with a minimal number of features. Data re-sampling technique such as SMOTE was found to be very productive to generate synthetic data observations for minority classes like CWS. The research-based historical data of developmental stuttering is a precious treasure. On such data, predictive modeling can be applied to explore hidden patterns and the causation-and-effect relationships between data variables concerning stuttering. The outcomes of this study may benefit several stakeholders like Researchers, Therapists, SLPs, and Consultants who offer professional services for the advancements in the area of fluency.