Scholarlink

Background <p>Prospective university students are regarded as highly vulnerable to psychological issues, including suicide. Despite the complexity of suicidal behaviors, innovative methodologies like Geographic Information System (GIS) mapping and Machine Learning have not been fully explored for predictive modeling and risk assessment. This study aims to investigate the prevalence and risk factors associated with suicidal behavior, offering a thorough understanding of the spatial distribution and predictive factors of suicidality.</p> Methods <p>Data from 1,485 participants were collected on socio-demographic characteristics, admission-related variables, health behaviors, and familial factors. Logistic regression analysis identified significant risk factors, while Machine Learning algorithms, including CatBoost and K-Nearest Neighbors, were used for prediction.</p> Results <p>The findings revealed a 20.5% prevalence of suicidal thoughts, with disparities across demographics and behaviors. Female participants, rural dwellers, and those from joint families exhibited higher suicidality rates. Repeat test-takers, academically struggling students, and those not coached professionally displayed elevated risks. Moreover, substance use, mental health issues, and family mental health and suicide history increased odds of suicidal behavior. GIS mapping identified regional variations, notably in the Sylhet division and Chittagong Hill Tracts. While, Machine Learning models were used to predict suicidal thoughts, with depression status as the most influential factor. Among all models, CatBoost achieved the best overall performance, with the lowest log loss, highest AUC, and strongest 95% confidence interval. KNN also performed well in accuracy, precision, and F1-score, but showed a slightly higher log loss, making CatBoost the most reliable model for predicting suicidal thoughts.</p> Conclusions <p>This study emphasizes the multifaceted nature of suicidal behavior, emphasizing the need for targeted interventions and support services to address mental health challenges and prevent suicide in this vulnerable population.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploring suicidal thoughts among prospective university students: a study with applications of machine learning and GIS techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月1日

查看全文
  • Mohammed A. Mamun,
  • Firoj Al-Mamun,
  • Md Emran Hasan,
  • Nitai Roy,
  • Moneerah Mohammad ALmerab,
  • David Gozal,
  • Md. Shakhaoat Hossain

摘要

Background

Prospective university students are regarded as highly vulnerable to psychological issues, including suicide. Despite the complexity of suicidal behaviors, innovative methodologies like Geographic Information System (GIS) mapping and Machine Learning have not been fully explored for predictive modeling and risk assessment. This study aims to investigate the prevalence and risk factors associated with suicidal behavior, offering a thorough understanding of the spatial distribution and predictive factors of suicidality.

Methods

Data from 1,485 participants were collected on socio-demographic characteristics, admission-related variables, health behaviors, and familial factors. Logistic regression analysis identified significant risk factors, while Machine Learning algorithms, including CatBoost and K-Nearest Neighbors, were used for prediction.

Results

The findings revealed a 20.5% prevalence of suicidal thoughts, with disparities across demographics and behaviors. Female participants, rural dwellers, and those from joint families exhibited higher suicidality rates. Repeat test-takers, academically struggling students, and those not coached professionally displayed elevated risks. Moreover, substance use, mental health issues, and family mental health and suicide history increased odds of suicidal behavior. GIS mapping identified regional variations, notably in the Sylhet division and Chittagong Hill Tracts. While, Machine Learning models were used to predict suicidal thoughts, with depression status as the most influential factor. Among all models, CatBoost achieved the best overall performance, with the lowest log loss, highest AUC, and strongest 95% confidence interval. KNN also performed well in accuracy, precision, and F1-score, but showed a slightly higher log loss, making CatBoost the most reliable model for predicting suicidal thoughts.

Conclusions

This study emphasizes the multifaceted nature of suicidal behavior, emphasizing the need for targeted interventions and support services to address mental health challenges and prevent suicide in this vulnerable population.