| dc.description.abstract | The study focuses on attention deficit hyperactivity disorder that is underdiagnosed in
Bangladeshi children, a critical public health concern. Due to the lack of awareness,
high levels of social stigma and a severe shortage of available and culturally-appropriate screening methods, many children with signs of ADHD are undiagnosed, thus impeding their behavioral and academic progress. This research aims to bridge this gap by developing a culturally sensitive, economical and data-driven machine learning algorithm to detect signs of ADHD in the early stages, and to develop an app interface of ADHD screening (SHUCHOK) in the Bangladeshi setting. Our method included a collection of a survey-based data set based on 110 participants between the ages of 4 and 17 with diagnosis of ADHD (n = 44) and without (n = 66). A DSM-5-based questionnaire translated into Bengali was used to capture behavioral symptoms, demographics, developmental history, and environmental factors. The four machine learning models were Random Forest, Support Vector Machine, Logistic Regression and Decision Tree which were trained and evaluated after subjecting the data
to extensive data preprocessing and feature selection approaches, including Recursive
Feature Elimination and Select Percentile. The Random Forest model had the highest accuracy (91%), precision (0.90), recall (0.88), F1-Score (0.90) and Area Under the Curve (0.95) when compared to the others. In this sample, the most important predictive characteristics of ADHD are inattention, fidgeting, study focus and level of irritation of the child. The main implication of this research is that it has great potential to enhance public awareness and early diagnosis of ADHD in low-resource countries such as Bangladesh. This study will significantly improve the accessibility of diagnoses and timeliness of interventions by offering a machine learning-based screening tool, which is proven to enhance the developmental outcome of the children affected. | en_US |
