Problematic internet use among children and adolescents is an escalating concern, often associated with mental health issues such as depression and anxiety. The severity of this issue can be classified into four levels. Current assessment methods are complex, requiring professional expertise, which limits accessibility. In this project, I address this issue by developing a machine learning framework, starting with data preprocessing and feature selection for multi-class classification. I initially apply decision tree and random forest algorithms, tuning hyperparameters to achieve optimal predictive performance and identifying the model best suited to this dataset. To further improve accuracy, I developed a custom neural network classifier and applied a stacking approach to combine multiple models, resulting in a refined model with superior precision. Results indicate that the stacking method outperforms decision tree algorithms in predicting levels of problematic internet use based on children’s physical activity and health data.
PRESENT RESEARCH GAP
However, traditional machine learning models face significant challenges when dealing with high-dimensional data, often struggling with overfitting and inefficiency. Existing base models are not well-suited for feature-rich datasets like the HBN dataset, making it difficult to accurately predict mental health outcomes. While recent research has explored machine learning applications in mental health, there remains a lack of models that demonstrate strong predictive performance in identifying mental health risks related to problematic internet use among adolescents.
Reference:
Rohitash Chandra and Ritij Saini. Biden vs Trump: Modeling US general elections using BERT language model. IEEE Access, 9:128494–128505, 2021.
To address this gap, this project focuses on developing a multi-class predictive model optimized specifically for the HBN dataset. By analyzing children’s physical activity and fitness data, the model aims to detect early indicators of problematic internet use, enabling timely interventions to promote healthier digital habits. Given the urgent need for reliable tools to assess and mitigate harmful internet use among adolescents, this project represents a significant step forward, offering an effective approach for early detection and proactive intervention.