Authors: G. Sudha Sadasivam; Mutyala Subrahmanyam; Dasaraju Himachalam; Bhanu Prasad Pinnamaneni; S. Maha Lakshme
Addresses: Department of CSE, PSG College of Technology, Coimbatore – 641 004, India ' Commerce Department, Sri Venkateswara University, Tirupati – 517 504, India ' Commerce Department, Sri Venkateswara University, Tirupati – 517 504, India ' Computer Science and Engineering, Rajalakshmi Engineering College, Chennai – 602 105, India ' Department of CSE, PSG College of Technology, Coimbatore – 641 004, India
Abstract: Financial reports of corporations publicise their performance. This in turn motivates manipulation of financial statements. Falsification of financial statements over prolonged period results in sudden collapse of multi-national companies, long-term economic loss to government and loss of trust of public. Detecting management frauds using normal audit procedures is time expensive as huge volume of data needs to be analysed. Hence additional analytical procedures should be used. The proposed work aims at automated analysis of annual reports using MapReduce paradigm to identify fraudulent companies. Annual reports of companies from public repositories are parsed to extract features for preparing a score card. Principal component analysis is applied on the score card to extract the principal features to train support vector machine. Experimental results show that 90% accuracy can be achieved using 10% to 25% of the principal features. Using MapReduce paradigm for feature extraction and classification improves the time efficiency by 85%.
Keywords: big data analytics; corporate governance; machine learning; financial statements; financial fraud; fraud detection; annual reports; financial statement manipulation; principal component analysis; PCA; feature extraction; support vector machines; SVM; classification; white collar crime.
International Journal of Big Data Intelligence, 2016 Vol.3 No.1, pp.51 - 60
Received: 27 Jun 2014
Accepted: 16 Jan 2015
Published online: 29 Dec 2015 *