Please use this identifier to cite or link to this item:
https://ptsldigital.ukm.my/jspui/handle/123456789/475659Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | Abdul Razak Hamdan, Prof. | - |
| dc.contributor.author | Massoud Vadoodparast (P56113) | - |
| dc.date.accessioned | 2023-10-05T06:40:47Z | - |
| dc.date.available | 2023-10-05T06:40:47Z | - |
| dc.date.issued | 2016-02-29 | - |
| dc.identifier.other | ukmvital:82220 | - |
| dc.identifier.uri | https://ptsldigital.ukm.my/jspui/handle/123456789/475659 | - |
| dc.description | Today’s detecting and preventing fraudulent financial transactions especially in credit card transactions from huge volume of data and make sure about accuracy of these data are playing important role in the banking and financial institutions business. Many researches have used data mining algorithms to detect fraudulent transactions. Normally more than one million transactions are created daily, so first issue is detection process in optimal way, is a time consuming process and mostly is done offline in static operation, usually the batch processing is used in specific period like daily, weekly or monthly to discover the fraud. It means preventing of happening fraudulent transaction does not occur in transaction time. The second issue is the learning machine or supervised algorithm like classification relies on accurate identification of fraudulent and non-fraudulent transactions, however these information usually do not exists or limited or the system using predefined rules and scenarios or static model and needs to update periodically. In order to fill these gaps and perform optimally, lead us to work in this area to optimize presented models. Thus, the research objectives are to propose optimized hybrid model and mechanism to cover these two issues better. Firstly, reduce huge data processing for transaction inspection and save the time and Secondly, detecting abnormality in customer behavior without having fixed or predefined scenarios or rules, and update customer behavior model once new transaction occurred by respective customer. In this research, the standard data mining methodology is adopted. About 40 million real data are collected form one bank and after preprocessing, six parameters selected for processing as: Purchase Transaction Type, Merchant ID, Media Type, Transaction Amount, customer identification and Transaction Date &Time. Toward this goal, K-MEANS, DBSCAN and AGGLOMERATIVE techniques employed that all are clustering techniques. Voted prediction mechanism from these three techniques is used. In proposed model, we are looking for local outlier in cluster with few members by using K-MEANS, high Local Outlier Factor (LOF) values by using DBSCAN or a single node in the tree by using AGGLOMERATIVE, for each customer separately. If two or more techniques vote the new transaction could be fraud the system consider it as fraud and raise the alert. We named it KDA hybrid model. We used RapidMiner software as modeling tool and developed software to test and extract fraudulent transaction from collected data. This model used in DSS system in order to help and advice users to make better decision regarding active transaction. As first result, the KDA model showed that by adding one more step before data processing that extracting only customer previous 100 transactions plus new transaction and process them, it means maximum 101 transaction inspection, instead of processing all transactions exist in database, can reduce and optimize the processing time significantly. Secondly, using clustering techniques, helps to avoid learning dataset necessity and overcome the static rules and predefined scenarios, therefore no need to update the model periodically, it update itself automatically. For evaluating KDA hybrid model 32 genuine fraud cases have used as external model evaluation. The False-Negative rates that show fraudulent transactions and model detect fraud was 81.25% for historical dataset in batch processing mode and 68.75% in online mode as successful detection for 1015 customers and 3,609,618 transactions. It clearly shows that this model can help to detect fraudulent transactions in both offline and online mode effectively. FDS&DSS software propose a good supporting procedure to detect fraudulent transaction. The FDS has developed by Microsoft Visual Studio 2010(VB.Net) and Microsoft SQL SERVER 2008 used as database engine,Certification of Master's/Doctoral Thesis" is not available | - |
| dc.language.iso | eng | - |
| dc.publisher | UKM, Bangi | - |
| dc.relation | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | - |
| dc.rights | UKM | - |
| dc.subject | Fraudulent transactions | - |
| dc.subject | Detection | - |
| dc.subject | Hybrid model | - |
| dc.subject | Dissertations, Academic -- Malaysia | - |
| dc.title | Fraudulent electronic transaction detection using hybrid approach of clustering techniques | - |
| dc.type | Theses | - |
| dc.format.pages | 172 | - |
| Appears in Collections: | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.