Random Forests with Perceptron Artificial Neural Networks Employed in Fraud Detection

• Grant number: 17/02859-0
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: August 01, 2017
• End date: July 31, 2019
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: André Carlos Ponce de Leon Ferreira de Carvalho
• Grantee: Luciano Carli Moreira de Andrade
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 13/07375-0 - CeMEAI - Center for Mathematical Sciences Applied to Industry, AP.CEPID

Abstract

Random Forests is a collection of Decision Trees that have been generated, each one in a slightly different way, using a random set of data. Decision trees are a methos to be guided through a path to a decision, which can be a simple or a complex decision of many values. Decision Trees are hierarchically branched structures and they guide a decision-making based on questions made in a particular sequence. They are one of the most commonly used classification techniques, because they are very efficient and easy to understand and use. Their classification accuracy is competitive compared to other methods. Decision Trees can generate reliability from some test instances that can then be applied to a broad population. These characteristics can be combined with those of Artificial Neural Networks (ANNs), such as adaptive, learning and generalization skills, allowing them to deal with inaccurate data and situations that are not fully defined. The proposal of this project is to investigate the use of Random Forests combined with ANNs to detect fraud. In order to protect organizations from abusive practices and criminal activities is necessary to combat fraud. The support of information technology is essential and techniques of electronic fraud detection systems are developed in various industries. In the insurance, telecommunications and financial sector, particularly the credit card industry, fraud detection is a vital aspect. The characteristics of support for decision-making of Random Forests combined with the learning and generalization characteristics of ANNs can allow the identification of frauds, mitigating the financial impacts that cause enormous losses and jeopardize the financial health and welfare of its beneficiaries.