Research of Classifier Threshold and Expected Risk of Mobile Phone Replacement Based on the Misclassification Cost

Abstract

Abstract:

The traditional classification algorithm treats the correct prediction and the error prediction equally, ignores the subjective factors and can't control the error rate well. Based on the users' consumption data selected from Xi'an branch of a mobile communication company, this paper studies the threshold and expected risk of forecasting mobile phone replacement by using Logistic model with misclassification cost. We find that:the Logistic model with misclassification cost has a good classification effect; different misclassification costs correspond to different optimal thresholds, but the prediction accuracy is basically the same; classification with a traditional threshold of 0.5 not only reduces the accuracy of the forecast but also increases the expected risk; the greater the difference in classification costs between positive and negative categories, the higher the expected risk for the classifier to predict; there is a dynamic equilibrium and mutual restraint between the optimal classifier's value, the optimal threshold and the expected risk. Thus, these results not only provide a multi-dimensional analysis framework for data mining researchers, but also provide a decision-making reference for manufacturers and vendors.

Key words: misclassification cost, algorithms, mobile users, threshold

Wang Chaofa, Sun Jingchun. Research of Classifier Threshold and Expected Risk of Mobile Phone Replacement Based on the Misclassification Cost[J]. , 2018, 30(12): 122-130.

References

[1] David E. C., Edward L., Hannan C. W. Predicting Risk-Adjusted Mortality for Trauma Patients:Logistic Versus Multilevel Logistic Models[J]. Journal of the American College of Surgeons, 2010,211(2):224-231
[2] Farquad M., Indranil B. Preprocessing Unbalanced Data Using Support Vector Machine[J]. Decision Support Systems, 2012,53(1):226-233
[3] Zhou T., Shan H., Banerjee A. Kernelized Probabilistic Matrix Factorization:Exploiting Graphs and Side Information[C]. Proceedings of the 12th SIAM International Conference on Data Mining. Philadelphia:SIAM, 2012
[4] Miwa S., Hirai T., Sumi K. Robust Face Detection Using One-classestimation and Real AdaBoost[J]. Electronics and Communications in Japan, 2014,97(7):39-47
[5] Mierswa I. Controlling Overfitting with Multi-objective Support Vector Machine[C]. Proceedings of the 9th Annual Conference on Genetic and Evolutionary Computation. In GECCO, 2007
[6] Fanyong C., Jing Z., Cuihong W., et al. Large Cost-sensitive Margin Distribution Machine for Imbalanced Data Classification[J]. Neurocomputing, 2017,224(8):45-57
[7] Fanyong C., Jing Z., Cuihong W. Cost-Sensitive Large Margin Distribution Machine for Classification of Iimbalanced data[J]. Pattern Recognition Letters, 2016,80(1):107-112
[8] Cuihuan D., Hong Z., Liming L., et al. Face Detection in Video Based on AdaBoost Algorithm and Skin Model[J]. The Journal of China Universities of Posts and Telecommunications, 2013,20(1):6-24
[9] 张婷婷,贺昌政,肖进. 基于动态分类器集成选择的不完整数据客户分类方法实证研究[J]. 管理评论, 2012,24(6):83-123
[10] 赵宇,黄思明,陈锐. 数据分类中的特征选择算法研究[J]. 中国管理科学, 2013,21(6):38-46
[11] 王雷,王欣,赵秋红. 基于和声搜索算法优化支持向量机的突发暴恐事件分级研究[J]. 管理评论, 2016,28(8):125-132
[12] 商丽媛,谭清美. 基于支持向量机的突发事件分级研究[J]. 管理工程学报, 2014,28(1):119-123
[13] 应维云. 随机森林方法及其在客户流失预测中的应用研究[J]. 管理评论, 2012,24(2):140-145
[14] 邹鹏,莫佳卉,江亦华,等. 基于代价敏感决策树的客户价值细分[J]. 管理科学, 2011,24(2):20-29
[15] Marc G. LogitBoost Autoregressive Networks[J]. Computational Statistics and Data Analysis, 2017,112(3):88-98
[16] Cao P., Zhao D. Z., Zaiane O. An Optimized Cost-sensitive SVM for Imbalanced Data Learning[C]. Advances in Knowledge Discovery and Data Mining Lecture Notes in Computer Science. Berlin:Springer, 2013
[17] Guoqing Z., Huaijiang S., Zexuan J., et al. Cost-sensitive Dictionary Learning for Face Recognition[J]. Pattern Recognition, 2016,60(7):613-629
[18] Cao P., Zhao D. Z., Zaiane O. A PSO-based Cost-sensitive Neural Network for Imbalanced Data Classification[C]. Trends and Appli-canons in Knowledge Discovery and Data Mining Lecture Notes in Computer Science. Berlin:Springer, 2013
[19] Jingjing Y., Lili T. Joint Inference about Sensitivity and Specificity at the Optimal Cut-off Point Associated with Youden Index[J]. Computational Statistics and Data Analysis, 2014,77(14):1-13