<![CDATA[The number of financial transactions has the potential to cause many violations of the law (fraud). Conventional machine learning has been widely used, including logistic regression, random forest, and gradient boosted. However, the machine learning can work as long as the dataset contains fraud. Many new financial technology companies need to anticipate the potential for fraud, which they have not experienced much. This potential for a crime can also be experienced by old service providers with a low frequency of previous fraud. With the data imbalance, traditional machine learningis likely to produce false negatives so that they do not accurately predict potential fraud. This study optimizes the machine learning approach based on Neural Networks to improve model accuracy through the integration of KNIME and Python Programming with KERAS and TensorFlow models. The study also conducts a comparative analysis to scrutinize the performance of Adam and Adamax Optimizer. Using data from European cardholders in 2013, this study proves that workflows and neural network algorithms can detect with up to 95% accuracy even with a very small fraud sample of only 0.17% or 492 of 284,807 transactions. In addition, the Adam optimizer performs higher accuracy than the Adamax optimizer. The implication is that this supervisory technology innovation can be developed to minimize transaction crimes in the financial services sector.]]>
Copyrights © 2022
