Financial distress and then the consequent failure of a business is usually an extremely costly and disruptive event. Statistical financial distress prediction models attempt to predict whether a business will experience financial distress in the future. Discriminant analysis and logistic regression have been the most popular approaches, but there is also a large number of alternative cutting - edge data mining techniques that can be used. In this paper, a semi-parametric Cox survival analysis model and non-parametric CART decision trees have been applied to financial distress prediction and compared with each other as well as the most popular approaches. This analysis is done over a variety of cost ratios (Type I Error cost: Type II Error cost) and prediction intervals as these differ depending on the situation. The results show that decision trees and survival analysis models have good prediction accuracy that justifies their use and supports further investigation.
|Title of host publication||Eleventh International Conference on Communication Networks, ICCN 2015/India eleventh International Conference on Data Mining and Warehousing ICDMW 2015/India eleventh International Conference on Image and Signal Processing, ICISP 2015|
|Editors||PD Shenoy, SS Iyengar, KB Raja, KR Venugopal, R Buyya, LM Patnaik|
|Number of pages||9|
|Publication status||Published - 1 Jan 2015|
|Event||11th International Conference on Data Mining and Warehousing (ICDMW) - Bangalore, Bangalore, India|
Duration: 21 Aug 2015 → 23 Aug 2015
|Name||Procedia Computer Science|
|Conference||11th International Conference on Data Mining and Warehousing (ICDMW)|
|Period||21/08/15 → 23/08/15|
FingerprintDive into the research topics of 'Predicting financial distress: A comparison of survival analysis and decision tree techniques'. Together they form a unique fingerprint.
Financial statement fraud detection using supervised learning methodsAuthor: Gepp, A., 10 Oct 2015
Supervisor: Kumar, K. (Supervisor) & Bhattacharya, S. (External person) (Supervisor)
Student thesis: Doctoral ThesisFile