论文部分内容阅读
认知诊断评估的主要问题是如何准确进行被试分类和项目属性标定。本文使用概率神经网络(PNN)和支持向量机(SVM)进行被试分类和属性标定,重点讨论PNN用于诊断的理论根据。模拟研究表明:PNN方法表现最好,训练速度快且具有很好判准率和标定准确率;PNN与GDD方法在分类上表现相当,在独立结构下PNN更好;线性SVM具有较好判准率和标定准确率。软计算中此类方法可非常方便推广至多级评分测验数据分析。
The main problem of cognitive assessment is how to classify the subjects and calibrate the attributes of the project accurately. In this paper, probabilistic neural network (PNN) and support vector machine (SVM) were used to classify and attribute the subjects, focusing on the theoretical basis of PNN for diagnosis. The simulation results show that the PNN method performs best, has fast training speed and good accuracy and calibration accuracy. The PNN and GDD methods perform fairly well in classification, and the PNN is better under independent structures. The linear SVM has a better criterion Rate and calibration accuracy. Such methods in soft computing can be easily extended to multistage scoring data analysis.