Design of A Novel Generative Adversarial Network for Outlier Prediction with AMBO Algorithm

Outlier detection identifies the objects which are deviated from the other objects in the dataset. However, outlier detection is challenged due to data scarcity and abnormal system behaviour. These limitations are resolved with Generative Adversarial Networks (GANs) which has significant attention with the capacity of new data generation. In this paper, the novel outlier detection approach Adaptive Mine Blast Optimization (AMBO) GAN is proposed for differentiating the fake data from the original one. GAN improves the data quality and efficiency by correcting errors and reducing the noise. GAN involves iterative backup loop between generator and descriminator. The network loss of GAN is minimized with AMBO algorithm which outperforms other metaheuristic approaches in terms of convergence and solution. In AMBO, the function values are minimized to near optimal solution during early iterations. The proposed approach is analysed with abalone age dataset in which the classification related metrics are considered for performance evaluation. In addition to that, the sensitivity, specificity, accuracy, AUC, and G-mean performance obtained with the proposed approach is 95%, 96%, 98%, 0.92, and 0.9564 respectively. The numerical outcomes show that the model is best suited for real-time environment and outperforms the functionality of the existing systems. Proposed appraoch is appropriate for real-time applications as it learns the dataset's internal data representations and complex data.