Improving Credit Card Fraud Detection with Class Imbalance Resilience and Dynamic Machine Learning Approaches

• Published in: 2024 5th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI) pp. 52 - 58
• Main Authors: Pamidi, Vyshnavi, Yellamma, Pachipala, Prasanth, Balisetty, T, Charmi Padmaja
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18-01-2024
• Subjects: Area Under the Precision-Recall Curve (AUPRC); Computational modeling; Credit cards; Exploratory Data Analysis (EDA); Fraud; Fraud Prevention; Heuristic algorithms; Imbalanced Data; Libraries; Machine learning algorithms; NumPy; Precision -Recall Curve; Random Forest Classifier; Scikit-learn; Security; SMOTE
• DOI: 10.1109/ICMCSI61536.2024.00014

This endeavor, which centers on the "Credit Card Fraud Detection Using Machine Learning in Python", is the main goal, and the aims to carefully build a trustworthy credit card fraud detection technology that reduces false positives, but it also follows a uniform formatting style and arranging and acknowledge the vital significance of credit cards. financial firms' efforts to prevent fraud and use machines. Educational tools, like NumPy and scikit-learn, to construct a binary classifier that adheres to specified style and guidelines for formatting. The dataset in question includes 31 parameters, out of which 28 are PCA-transformed components, and attributes such as "Amount," "Time," and "Class" are emphasized in dependable styling. Because of the dataset's notable class imbalance, the proposed work is using AUPRC for accuracy evaluation to uphold uniformity with selected methodology and style. The concept makes use of several crucial tools and libraries, including as imblearn, Matplotlib, and Python and itertools/collections, which are crucial elements for making certain that method for detecting credit card fraud is both fashionable and functional. Three steps are included in the development process. essential actions: carrying out regular EDA in accordance with selected style, using a variety of equipment learning methods that correspond with the components the proposed work chosen, and in the end, developing, assessing, and choosing the optimal model to incorporate into system for detecting fraud, guaranteeing a regular and precise result.