Browsing by Author "Mahmoudi, Nader"
Now showing 1 - 2 of 2
- Results Per Page
- Sort Options
ArticlePublication Metadata only Detecting credit card fraud by modified Fisher discriminant analysis(Elsevier, 01.04.2015) Mahmoudi, Nader; Duman, Ekrem; Industrial Engineering; DUMAN, Ekrem; Mahmoudi, NaderIn parallel to the increase in the number of credit card transactions, the financial losses due to fraud have also increased. Thus, the popularity of credit card fraud detection has been increased both for academicians and banks. Many supervised learning methods were introduced in credit card fraud literature some of which bears quite complex algorithms. As compared to complex algorithms which somehow over-fit the dataset they are built on, one can expect simpler algorithms may show a more robust performance on a range of datasets. Although, linear discriminant functions are less complex classifiers and can work on high-dimensional problems like credit card fraud detection, they did not receive considerable attention so far. This study investigates a linear discriminant, called Fisher Discriminant Function for the first time in credit card fraud detection problem. On the other hand, in this and some other domains, cost of false negatives is very higher than false positives and is different for each transaction. Thus, it is necessary to develop classification methods which are biased toward the most important instances. To cope for this, a Modified Fisher Discriminant Function is proposed in this study which makes the traditional function more sensitive to the important instances. This way, the profit that can be obtained from a fraud/legitimate classifier is maximized. Experimental results confirm that Modified Fisher Discriminant could eventuate more profit.Master ThesisPublication Embargo Profit-oriented classification : new approaches and business applications(2015-11) Mahmoudi, Nader; Duman, Ekrem; Duman, Ekrem; Öztop, Erhan; Ağaoğlu, M.; Department of Industrial Engineering; Mahmoudi, NaderClassification problems are the most common prediction problems that have traditionally been tackled by the data mining (DM) algorithms. The objective taken in these algorithms is a statistical one aimed to minimize the number (or, the weighted number) of incorrectly classified observations (instances). Recently, the cost-sensitive classification got researchers' attentions as the existing algorithms are not able to deal with special concerns in some popular problems. There are two main special concerns. The first one is the case when the number of observations varies in different classes - called class imbalance (or skewness). The second issue is the case when there are naturally different costs of misclassification that should be considered while implementing a classification algorithm. This study includes two types of profit-oriented approaches to deal with four real-life problems. Firstly, we have modified Fisher Discriminant Analysis (FDA) converting it to a profit-sensitive approach. The Profit-sensitive Fisher Discriminant Analysis (PFDA) modifies the existing error-based FDA in a way that puts emphasize on the profitable observations inheriting the main error-minimization assumptions. The second approach called profit-based modeling tries to classify the observations with regard to total net profit rather than errors in classification. This approach searches solution space for a discriminating function with maximum net profit using meta-heuristics. Four meta-heuristics are utilized to implement the profit-based classification approach including Migrating Birds Optimization (MBO), Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA). In this study, we also have proposed a modification on MBO (m-MBO). Results show that the profit-sensitive FDA could catch profitable positives more than its original version however, it has less number of positives correctly classified. The profit-sensitive approach adopts the error-minimization assumptions such that the priority in classification is set for profitable observations. On the other hand, the profit-based approach could reach more profit-making solutions using an objective function of maximizing the total net profit. As this approach totally neglects error-minimization assumptions while training, it showed under-performance in true positive rate. Among the meta-heuristics utilized in profit-based approach, the Artificial Bee Colony (ABC) and modified version of MBO (m-MBO) always outperform other meta-heuristics.