大数据背景下ROC曲线介绍与应用

Introduction and Application of ROC Curve under the Background of Big Data

ROC曲线是研究二分类问题中一种简便、高效的综合性工具,它通过连接真阳率与假阳率在不同切断点或者阈值下的值,构造出一条单调递增曲线,曲线下的面积AUC即可作为衡量诊断效果的指标,面积数值越大,则代表该分类方法越有效。ROC曲线在算法效度研究,心理测评,[1]临床医学诊断中都有广泛的应用。本文简单介绍了ROC曲线的构造方法、基本性质和实际意义,并以生物医学和机器学习的三个例子深入介绍了不同情形下ROC曲线的应用,最后总结ROC曲线的优缺点和应用场景。

ROC curve is a simple and efficient comprehensive tool in the study of binary classification. It constructs a monotonic increasing curve by connecting the true positive rate and false positive rate at different cut-off points or thresholds. The area AUC under the curve can be used as an index to measure the diagnosis effect. The larger the area is, the more effective the classification method is. ROC curve is widely used in algorithm validity research,psychological evaluation, clinical diagnosis. This paper briefly introduces the construction method, basic properties and practical significance of ROC curve, and introduces the application of ROC curve in different situations with three examples of biomedicine and machine learning. Finally, it summarizes the advantages and disadvantages of ROC curve and application scenarios.