Towards Interpretable Anomaly Detection: Unsupervised Deep Neural Network Approach using Feedback Loop