Comparative analysis of pretrained deep neural networks for anomalies detection

Robust anomaly detection algorithm is critical to a significant number of applications. Deep Neural Network can resolve this problem with the good accuracy. This paper demonstrates the ability of solving anomalies detection using pretrained convolutional neural networks. Three pretrained neural networks were used to solve this problem - SqueezeNet, GoogleNet and Resnet18. The contribution of this paper consists in a comparative analysis of performances of these deep neural networks. The result was compared against the following metrics: accuracy, recall, precision and the F1 Score. The neural networks were tested against two different public available datasets, showing different type of anomalies.