In the last two to three decades, use of credit cards is increasing rapidly due to fast economic growth in developing countries and worldwide globalization issues. Financial institutions like banks are facing a very tough time due to fast-rising cases of credit card loan payment defaulters. The banking institution is constantly searching for the perfect mechanisms or methods to identify possible defaulters among the whole set of credit card users. In this chapter, the most important features of a credit card holder are identified from a considerably large set of features using metaheuristic algorithms. In this work, a standard data set archived in UCI repository of credit card payments of Taiwan is used. Metaheuristic algorithms like particle swarm optimization, ant colony optimization, and simulated annealing are used to identify the significant sets of features from the given data set. Support vector machine classifier is used to identify the class in this two-class (loan defaulter or not) problem.
TopIntroduction
Credit card refers to a card with a magnetic strip and/or a microchip ensuring its uniqueness. Credit cards are issued by banks, credit providers and other financial institutions that allow the card holder to have a loan of funds. These funds can be used to pay for products/services as per necessity of the card holder. The credit cards are issued based on the state that the card holder will pay the loan amount and any extra charges that have been mutually decided. The loan provider institution also grants a line of credit (LOC) to the cardholder. This facility allows the consumer to borrow loan in the form of cash. The issuer can set maximum limit of loan depending upon the credit rating. Generally credit card transactions are categorized under two heads:
In case of POS Transaction, the card holder may purchase items as per his/her necessity and the bill is settled through credit card. The settlement involves the transfer of fund from bank to the merchant’s account and consequently the card holder completes his purchase. But the amount that the bank pays to the merchant on behalf of the hard holder is amount that the bank / credit provider company charges the card holder with in the credit card bill at the end of the month. As the cardholder pays back the billed amount he / she become eligible for the credit amount for the next month.
Obvious from the prior section, that huge number of transactions is being made through credit cards. Card holders generally have a specific pattern of transactions. Based on these patterns or deviation from these patterns lots of works have been proposed regarding fraud-detection in credit card transactions. For instance (Ingole, & Thool, 2013), (Jiang, et al. 2018), (Dhankhad, et al. 2018), (Raj, & Portia, 2011) (Kazemi, & Zarabi,) and (Pozzolo, et. al 2018) have mentioned contribution on fraud identification in credit card transaction.
Metaheuristic optimization technique is a state of the art searching method which can be used to optimize a function with different parameters (Wahono, et al. 2014). There are two type of searching method local and global search. A wide range of metaheuristic technique has emerged over the last few decades (Zhao, et al. 2016). Metaheuristic technique like Simulated Annealing (SA), Tabu Search (TS) and Iterated Local Search (ILS) fall under local search technique and Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Differential Evolution (DE) and Genetic Algorithms (GA) fall under global search technique.
In today’s world any financial and banking transaction is very sensitive in terms of risks involved due to uncertainty and volatility in the market. In this scenario identifying loan defaulter or credit card defaulter is very essential by the financial institution.
In this present work different features or parameters of portfolio dataset collected from UCI repository are optimized with metaheuristic method like SA, PSO and ACO and then the optimized feature set is used to classify the data with different supervised and unsupervised machine learning technique.
Most of the portfolio management datasets have multiple features which may lead to data over fitting and may result in inaccuracy in classification (Busetti, 2000). This is due to the concept of “curse of dimensionality”. This feature set can be optimized with metaheuristic algorithm which produces good result (Derigs, & Nickel, 2003; Jarraya, 2013). In this chapter the next two sections deal with literature survey and scope of the work. The third section is related to overview of different Meta-heuristic algorithms. Then a brief description is given on feature selection and optimization in the next section. Then dataset used and result and discussions are given. Finally in the conclusion the importance of the work is given.