Efficient loss updated XGBoost with deep emended genetic algorithm for detecting online fraudulent transactions.

In the fast-paced technological era, online financial transactions have gained widespread use as it offers significant merits to customers for easy transfer of money through smart phones. Nevertheless, fraudulent transactions put individual's money into risk, for which, suitable approaches are required to detect such deceits. Concurrently, with the progress of ML (Machine Learning) approaches, existing works have bidden to identify the fraudulent and normal transactions. However, studies lacked in accordance with accuracy rate and only limited focus has been provided for detection of generalized fraudulent transactions. Considering this, the current study considers IoT fraud dataset and proposes DEGA (Deep Emended Genetic Algorithm) to attain better performance for detecting fraudulent and normal transactions. This model employs a competitive approach, integrating, new crossover and selection methods. This intend to improvise the ability of global search and partition the chromosomes into losers and winners. This ensures high quality parent for selection. Besides, a dynamic-mutation function is also proposed for enhancing the model's searching ability. Subsequently, the study proposes EL-UXGB (Efficient Loss-Updated eXtreme Gradient Boosting) wherein dual sigmoid loss functions are proposed to resolve the imbalanced label cases. The overall performance of this study is assessed through analysis that confirms its effectiveness in detecting fraudulent transactions. [ABSTRACT FROM AUTHOR]