Paper Title
Machine Learning Techniques in Corporate Fraud Detection

Abstract
In a competitive environment, fraud is a business-critical problem, which can result in enormous losses and irrecoverable damages for an organization. Existing methods of fraud detection making substantial use of auditing is not only time consuming but is also inaccurate. There is a need to make use of computational methods for this fraud detection problem. The paper implements the applications of Machine Learning in corporate fraud detection. A comparative analysis is made between traditional machine learning algorithms such as K-Nearest Neighbors (KNN), Decision Trees, Adaboost and so on, and Deep Learning techniques using Artificial Neural Networks with error backpropagation. The performance is assessed based on evaluation metrics such as Accuracy and F1-Score. The results suggest that the performance of Artificial Neural Networks is far better than the traditional machine learning classifiers. The accuracy and F1-score of Neural Network was found to be 92% and 0.64 as against the KNN algorithm with accuracy of 87.9% and F1-score of 0.456. The impact of number of features on classifier performance is also analyzed and the best results are obtained for 25 features out of 41. This paper aims to assess what can go wrong in a corporate setting. It would be of great help to many organizations, and they can take corrective actions and prevent such events in the future. Keywords - Machine Learning, Neural Networks, Fraud Detection.