Children’s Drawings and Mental Health: A Machine Learning Approach

Mental health disorders among children and adolescents constitute a growing global public health concern, affecting about 14% of young people and leading to lasting social, educational, and developmental challenges (WHO, 2021). In low- and middle-income countries such as India, this burden is aggravated by limited mental health infrastructure, cultural stigma, and barriers to early identification. As evidenced by student absenteeism, school dropouts, and growing disinterest in learning, mental health profoundly influences human capital formation and future labor market outcomes. Yet, traditional screening tools—standardized questionnaires and clinical interviews— remain ill-suited for large-scale outreach or implementation in resource-constrained contexts. This is because they require trained professionals and high literacy levels or due to social desirability bias among respondents who may be reluctant to disclose emotional difficulties.
Children’s drawings provide a compelling alternative for mental health assessment, offering a non-verbal, culturally adaptable, and engaging form of emotional expression. Building on the theoretical underpinnings of the “draw a self-portrait” test, this study integrates psychological insights with machine learning methods to analyze visual and structural aspects of children’s drawings. These range from color usage and spatial organization to stroke density and symbolic representation—features that human enumerators might overlook or evaluate inconsistently.
Despite advances in art therapy and artificial intelligence in healthcare, there remains a notable gap in developing scalable and objective tools to assess children’s mental health rapidly and reliably. This study addresses that gap using data from 1,600 students aged 11–15 years, enrolled in 80 government schools across Rajasthan, India. Partnering with the Ministry of Education, both standardized mental health assessments and children’s drawings were collected and analyzed to train and validate machine learning models capable of predicting psychological wellbeing from drawing features.
The findings highlight the potential of computational approaches to complement traditional mental health screening, enabling early detection of at-risk children without the need for clinical expertise. From a policy perspective, this low-cost and non-intrusive method offers a practical way to triage students—distinguishing those requiring advanced care from those who may benefit from preventive support.
By combining economics, psychology, and computational science, this research advances the frontier of digital mental health assessment and contributes to global efforts to democratize mental health screening and access to care. The approach underscores how technology-driven, culturally sensitive tools can transform early intervention strategies and support the emotional wellbeing of vulnerable youth in developing contexts contributing towards both public science and policy.