FINGERPRINT REGISTRATION FRAUD DETECTION USING TIGHT CLUSTERING ON EMPLOYEE’S PRESENCE AND ACTIVITY DATA

• Main Author: Kamil, Irfan
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70698
• Institution: Institut Teknologi Bandung

Detecting fraud in fingerprint registration poses a unique challenge as we cannot rely on an existing employee’s attribute. Furthermore, analyzing using a supervised algorithm cannot handle unlabeled data that generated uniquely for this case. We study the patterns of employee’s presence and activity report data and found that fraud action tends to be closely similar to other fraud action. Therefore, we propose a tight clustering method to detect fraud in fingerprint data using Density-based spatial clustering of applications with noise algorithm (DBSCAN), as tight distance calculation removes non-fraud data because non-fraud data is generated to be unique naturally. This research proposes three methods to detect high likelihood between historical presence data and employee activity. These similarities tend to indicate fraud for employee presence and activity report because naturally those data should be generated dynamically. P2 methods with 3 days minimum distance shows highest F-Measure for 100%. But coverage for that method only 55%. To increase coverage to 100%, P2 methods with 5 days minimum distance or 7 days minimum distance may be used. For next research, activity research may be checked substantially as an indicator for presence fraud. Thus, those fraud indications also need to be investigated whether it is a fraud or just a coincidence. This investigation is required to measure accuracy for outlier detection method.