基于Cox回归的雷达操纵员认知能力智能化选拔研究
Research on intelligent selection of cognitive ability of radar operators based on Cox regression
马晓岩 1卢宏亮 1张志龙 1刘方霆 2郭璞 2尹海军 1朱霞1
作者信息
- 1. 空军军医大学军事医学心理学系军事心理学教研室,陕西 西安 710032
- 2. 解放军93897部队,陕西 西安 710082
- 折叠
摘要
目的 基于Cox回归探讨通过认知能力范式任务测评构建雷达操纵员智能化选拔模型.方法 单因素Cox回归纳入22 名雷达操纵专业新兵和25 名非专业新兵,选取以往选拔方法、智力水平以及空间N-back、三维1-back、精神运动警觉性任务(PVT)的反应时(RT)和正确率(ACC)为自变量,因变量以新兵在新训期间(120 d)首次连续 3 次专业考核成绩达到良好的日期作为时间变量.多变量Cox回归纳入 88 名雷达操纵专业新兵和49 名非专业新兵,以单因素Cox回归与时间变量显著相关(P<0.05)的因素确定自变量,因变量同单因素Cox回归.结果 以往选拔方法、3-back ACC、4-back ACC、三维1-back RT、PVT RT共5 项因素构成了最终的选拔模型(P<0.05).通过列线图呈现概率预测值,采用Bootstrap法进行验证得出第60、90、120 日达标率校准曲线显示选拔模型的预测概率与实际观察值吻合较好.通过受试者工作特征曲线对选拔模型预测效能进行评估,结果显示在第60、90、120日均展现出较好的鉴别能力,其中第60 日曲线下面积(AUC)为 0.73(P=0.01),95%CI为0.59～0.87,第90 日AUC 为 0.86(P=0.01),95%CI为 0.80～0.93,第 120 日 AUC 为 0.76(P=0.01),95%CI为0.68～0.84;以往选拔方法在第60 日AUC为0.55(P=0.44),95%CI为0.42～0.69,第90 日AUC为0.60(P=0.05),95%CI为0.51～0.69,第120 日AUC为0.61(P=0.02),95%CI为0.51～0.70;比较该选拔模型及以往选拔方法AUC之间的统计学差异结果显示第90 日及第120 日的AUC差异显著(P<0.05),表明该选拔模型的鉴别能力优于以往选拔方法.决策曲线分析法显示该选拔模型曲线绝大部分在以往选拔方法曲线的上方,且阈值宽度更广,表明相比以往选拔方法,用该选拔模型挑选雷达操纵专业兵会得到更大的净收益.结论 本研究构建的选拔模型对专业绩效情况反馈更加精确,能有效降低选拔风险与时间成本.
Abstract
Objective To explore the construction of intelligent selection model of radar operators by task evaluation of cognitive ability paradigm based on Cox regression.Methods Univariate Cox regression analysis included 22 radar operation professional recruits and 25 non-professional recruits.Past selection methods,intelligence level,reaction time(RT)and accuracy(ACC)of spatial N-back,three-dimensional 1-back and psychomotor vigilance task(PVT)were selected as independent variables.The dependent variable was the date when the recruits achieved good scores in three consecutive professional examinations for the first time during the training period(120 d).Multivariate Cox regression analysis included 88 radar operation professional recruits and 49 non-professional recruits.Independent variables were determined by univariate Cox regression which was significantly correlated with time variables(P<0.05),and dependent variables were the same as in the univariate Cox regression.Results The final selection model was composed of five factors:past selection methods,3-back ACC,4-back ACC,three-dimensional 1-back RT,and PVT RT(P<0.05).The predicted probability values were presented through a nomogram,and Bootstrap method was used to verify the calibration curves of good rate on the 60th day,the 90th day,and the 120th day,which showed that the predicted probability of the selection model was in good agreement with the actual observation value.The predictive efficiency of the selection model was evaluated by the receiver operating characteristic curve.The results showed that the discrimination ability was better on the 60th day,the 90th day,and the 120th day,where the area under the curve(AUC)on the 60th day was 0.73(P=0.01),and 95%CI was 0.59-0.87;the AUC on the 90th day was 0.86(P=0.01),and 95%CI was 0.80-0.93;and the AUC on the 120th day was 0.76(P=0.01),and 95%CI was 0.68-0.84.For past selection methods,the AUC on the 60th day was 0.55(P=0.44),and 95%CI was 0.42-0.69;the AUC on the 90th day was 0.60(P=0.05),and 95%CI was 0.51-0.69;and the AUC on the 120th day was 0.61(P=0.02),and 95%CI was 0.51-0.70.The statistical difference between the AUC of the selection model and the past selection methods showed that the AUC of the 90th day and the 120th day was significantly different(P<0.05),indicating that the discrimination ability of the selection model is better than that of the past selection methods.Decision curve analysis showed that most of the curve of the selection model was above the curve of the past selection methods,and the threshold width was wider,which indicated that the selection model could get greater net benefits compared with the past selection methods when selecting radar operation professional recruits.Conclusion The selection model constructed in this study is more accurate for the feedback of professional performance and can effectively reduce the selection risk and time cost.
关键词
智能化选拔/雷达操纵员/认知能力/Cox回归Key words
intelligent selection/radar operator/cognitive ability/Cox regression引用本文复制引用
基金项目
军委后勤保障部研究项目(22BJZ12)
军委后勤保障部研究项目(AKJWS221J001)
军委后勤保障部综合计划局"十三五"重大项目(AWS17J012)
出版年
2024
国家科技期刊平台
NSTL
万方数据