| Popis: |
Depression has become increasingly common among adolescents worldwide and has severe negative impact on an individual’s mental and social development. Early identification and diagnosis of depression during adolescence are crucial for improving the well-being of affected individuals. While natural language processing (NLP) has shown effectiveness in identifying depression in adults via social media, its application in predicting depression in adolescents remains challenging. This study aimed to develop a simple yet highly applicable method for predicting depression in adolescents within a school setting. We collected written compositions from 4,715 students aged 10 to 17, using their scores on the Children's Depression Inventory (CDI) to categorize them into high-risk and low-risk groups for depression. Then, we developed three types of computational models combining various feature extraction (theory-based vs. data-based) and classification techniques (classical machine learning algorithm vs. state-of-the-art deep neural networks), and compared their predictive performance in identifying individuals at risk. We found that all models exhibited promising performance in predicting depressive tendencies, with the recurrent neural network model outperforming the others. Our study demonstrates the feasibility of employing students’ written compositions to identify those at higher risk of depression, and providing a potential solution for early detection of depressive tendencies in adolescents. |
