Multi-Agent System And Distributed Data Mining Based Automated Credit Scoring System

The ability of a machine to perform functions that are normally associated with human intelligence, such as reasoning, planning, recognition, perception, cognition, learning, understanding, and problem-solving is one of the most interesting research fields in the computer science community. That defines the main challenge in artificial intelligence. Its major objective is to develop and use a machine to imitate some intellectual capabilities of human brain and to develop the related theories and techniques.The concept of an agent has become an important concept in AI. Intelligent agents are closely related to software agents, an autonomous software program that carries out tasks on behalf of users. In computer science, the term intelligent agent may be used to refer to a software agent that has some intelligence, learning methods and even regardless for none rational agents. These learning methods have proven to be one of great practical value in a variety of application domains, especially in data mining (also known as Knowledge Discovery in Database-KDD) which can be define as the process of extracting previously unknown, valid, and actionable information from large databases and then using the information to make crucial business decision.Actually, the marriage of agent technology and data mining is driven by challenges faced by both communities, and the need of developing more advanced intelligence system. Both, Agent technology and Data Mining, areas are currently very active.In our day, financial data are continuing to be produced and collected at an unprecedented rate. This is due to the increasing of economic globalization and evolution of information technology. That forced financial institutions to think about the need for automated approaches for an effective and efficient utilization of massive amount data in order to help in strategic planning and investment decision-making.In the same idea we’ll consider the credit scoring in banks institutions which have a challenge focused on the fact that they don’t know if a given customer will pay back or default on the loans that they grant them. They assess an investor’s risk of loss arising from a borrower who does not make payments as promised.In this thesis we combine multi agent system technology and distributed data mining techniques to deal with the above problem especially in automating the credit scoring for a optimal solution by building a predictive model from the historical data on customer information, for making a good credit approval decisions which can go up to minimize the probability of bankrupt of a company because the risk may be anticipated. This system will be a distributed system composed of autonomous sites that communicate with one processing site through a computer network.The proposed distributed approach deal with the agent mining interaction and integration to achieve the above goal. The site autonomy is justified by the fact that, the global predictive model is based on the merge and synchronization of multiple local models from each site. The processing site is in charge of the global predictive model building. In the proposed approach some challenges are identified. Those challenges are based on the agent coordination and communication mechanism, distributed data mining, time consuming and the best choice for algorithms.A design and implementation in a single site of ACSS is done and simulated using four technologies, JADE, Java, Java Data Mining and Oracle. Two mining algorithms are used, Decision Tree and Naive Bayes, to build the predictive model from a dataset provided by UCI Machine learning and repository for machine learning and intelligence system. The dataset had1000instances including300bad loans and700good loans.