Scholarlink

Background <p>Children and adolescents with Autism Spectrum Disorder (ASD) are at increased risk for suicidal ideation (SI) and suicide attempts (SA). This study used machine learning to predict SI and SA in this population and to identify the key features driving each model's predictions.</p> Methods <p>A cross-sectional study was conducted from May 2023 to December 2024 in Jordan, recruiting children and adolescents diagnosed with ASD. Participants underwent standardized assessments, and Recursive Feature Elimination (RFE) was used in R Studio to determine the top five predictors from a comprehensive list of psychiatric, psychological, and sociodemographic variables. Subsequently, machine learning algorithms, including logistic regression, random forest, XGBoost, and ensemble methods, were applied to predict SI and SA.</p> Results <p>Among children with ASD, random forest demonstrated the highest performance for predicting SI (area under the curve: 0.97), while logistic regression was most effective for SA (area under the curve: 0.99). For adolescents, logistic regression emerged as the best model for SI (area under curve: 0.99) and SA (area under curve: 0.94). The top three predictors for SI among children were anxious/depressed behavior, physical abuse, and bullying. In contrast, SA predictors included anxious/depressed behavior, family history of suicidal attempts, and aggressive behavior. For adolescents, the top predictors for SI were anxious/depressed behavior, somatic complaints, and physical abuse, while SA predictors included changes in routine, withdrawn/depressed behavior, and bullying.</p> Conclusions <p>Machine learning models with RFE show potential for enhancing suicide risk prediction in ASD populations, but further longitudinal studies are needed to validate their real-world applicability.</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

Using Machine Learning Algorithms in the Prediction of Suicidal Ideation and Attempts Among Children and Adolescents with Autism Spectrum Disorder

  • 期刊论文

  • 发表日期：

    2025年8月30日

查看全文
  • Mahmoud Gharaibeh,
  • Mohammad Nayef Ayasrah,
  • Ayoub Hamdan Al-Rousan,
  • Yusra Jadallah Abed Khasawneh,
  • Mohamad Ahmad Saleem Khasawneh

摘要

Background

Children and adolescents with Autism Spectrum Disorder (ASD) are at increased risk for suicidal ideation (SI) and suicide attempts (SA). This study used machine learning to predict SI and SA in this population and to identify the key features driving each model's predictions.

Methods

A cross-sectional study was conducted from May 2023 to December 2024 in Jordan, recruiting children and adolescents diagnosed with ASD. Participants underwent standardized assessments, and Recursive Feature Elimination (RFE) was used in R Studio to determine the top five predictors from a comprehensive list of psychiatric, psychological, and sociodemographic variables. Subsequently, machine learning algorithms, including logistic regression, random forest, XGBoost, and ensemble methods, were applied to predict SI and SA.

Results

Among children with ASD, random forest demonstrated the highest performance for predicting SI (area under the curve: 0.97), while logistic regression was most effective for SA (area under the curve: 0.99). For adolescents, logistic regression emerged as the best model for SI (area under curve: 0.99) and SA (area under curve: 0.94). The top three predictors for SI among children were anxious/depressed behavior, physical abuse, and bullying. In contrast, SA predictors included anxious/depressed behavior, family history of suicidal attempts, and aggressive behavior. For adolescents, the top predictors for SI were anxious/depressed behavior, somatic complaints, and physical abuse, while SA predictors included changes in routine, withdrawn/depressed behavior, and bullying.

Conclusions

Machine learning models with RFE show potential for enhancing suicide risk prediction in ASD populations, but further longitudinal studies are needed to validate their real-world applicability.