Machine Learning: Anomaly Detection in Practice

Anomaly Detection

Table of contents

Mission introduction

Unsupervised anomaly detection

data set

method

Whether autoencode can restore the original type of image, judge whether it is normal based on the reconstruction loss. The

reconstruction error is regarded as an abnormality score.

Evaluate

ROC and AUC scores are used for evaluation, and acc cannot be used for evaluation.

Baseline

Three kinds of autoencoder

You can add a classifier later, train a classifier, and use the output of the classifier as a probability distribution to represent the abnormality probability.

Report

Report evaluation criteria