Prototypical-Networks-Classification

This project introduces a supervised prototypical network for multiclass network attack detection, utilizing a neural network encoder with Euclidean distance-based classification to centroids. The approach models each class as a mean point in a multidimensional space, enabling precise classification. The system achieves an impressive accuracy of 90% with a mean false positive rate of 1.19%. The model architecture allows it to be visualized easily (using PCA) in a 3-dimensional space, making it easier to understand the classifications.