|
Abstract
|
The study of the impact of data preprocessing on fraud detection accuracy in deep learning models is critical in today's digital landscape, where financial transactions increasingly occur online. With the rise in online transactional activities, fraud has become a significant concern for both consumers and financial institutions. The necessity of effective fraud detection systems is underscored by the substantial losses incurred due to fraudulent activities. Data preprocessing plays a vital role in enhancing the accuracy of machine learning models used for fraud detection, as it helps to clean and structure data, making it more suitable for analysis and improving model performance [7].
Data preprocessing involves several key steps, including data cleaning, transformation, integration, and reduction. These processes are essential for handling issues such as missing values, outliers, and inconsistencies within the dataset. For instance, studies have shown that applying normalization and dimensionality reduction techniques like Principal Component Analysis (PCA) can significantly enhance the performance of classifiers used in fraud detection systems. By refining the dataset before training machine learning models, researchers have observed improved accuracy rates, with some classifiers achieving over 95% accuracy after preprocessing [7, 8]. The effectiveness of various machine learning algorithms in detecting fraudulent transactions is also influenced by the quality of the input data. Algorithms such as Logistic Regression, Naïve Bayes, and Random Forest have been widely employed in fraud detection tasks. However, their performance can vary significantly based on how well the data has been preprocessed. For example, studies indicate that models trained on preprocessed datasets consistently outperform those trained on raw data, highlighting the importance of preprocessing in achieving high detection rates [2].
Moreover, addressing class imbalance in datasets is another
|