CUSTOM FRAUD ANALYSIS USING DEEP LEARNING TECHNIQUES

Applications for deep learning have been widely adopted and proven beneficial in a number of sectors, including tax, finance, telecommunications, insurance, and aviation. Deep learning technology has been applied to fraud detection, especially by tax authorities. Actually, the most challenging problem facing modern revenue collection is tax fraud. Determining if deep learning techniques could be applied to tailored fraud detection was the aim of this study. In the data, a total of 13,999 records were preserved, of which particular data for the study was taken from Ethiopian Customs Commission. In total, 16 distinct features were used in the study. Once collected, the data is preprocessed and packed in a way that is suitable for deep learning tasks. Fraud detection is one part of total fraud management techniques that may be automated and helps to reduce the human portions of the screening/checking process. A trait like as fraud was used as a target class to create deep neural network models. Furthermore, the models' classification and detection ability, as well as the validity of the rules created using deep neural networks, were evaluated. The outcomes are compared in order to identify an acceptable model with improved precision and predictability. As a result, the deep neural network model after oversampling outperforms other models in terms of accuracy. The experimental results show that deep neural networks outperform previously recommended supervised machine learning techniques across all assessment criteria, with accuracy, precision, recall, and F-1 score of 99.6%%, 99.4%, 99.7%, and 99.6%, respectively.