【摘要】 目的 使用机器学习算法对症状自评量表(SCL-90)进行简化，在保证较高准确率的前提下，提高筛查效率。方法 在获得25 913名高中生有效数据的基础上，采用梯度提升回归树和随机森林算法建构分类模型，使用前向选择的方法选择特征，并通过十折交叉验证将简化后的量表的测验结果与验证。结果 梯度提升决策树的性能略优于随机森林，SCL-90可以由原来的90个题目减少为47个，各维度预测准确率均达到95%以上。结论 利用机器学习简化后的版本在保证准确率的基础上有效减少了题目数量，缩短了测试时间，有助于在学校测评情境中提升高中生心理健康筛查效率。更多还原

【Abstract】 Objective To simplify the Symptom Checklist-90(SCL-90) through machine learning algorithm, and to enhance screening efficiency on the premise of ensuring a high level of accuracy. Methods Classification models using gradient boosting regression tree and random forest algorithm were constructed on the basis of SCL-90 data from 25 913 high school students. Feature selection was carried out using forward selection, and the simplified scale’s test results were validated through ten-fold cross-validation. Results The performance of gradient boosting regression tree slightly outperformed random forest. The SCL-90 was reduced from its original 90 items to 47 items, with prediction accuracy exceeding 95% in all dimensions. Conclusion The simplified version utilizing machine learning algorithm maintains accuracy with the number of questions reduced and the testing time shortened. This enhancement contributes to improving the efficiency of mental health screening for high school students in school assessment scenarios.更多还原