I was researching about the many semi-supervised models that are there for anomaly detection. But none of them mentioned the ratio of labeled and un-labeled data that will be needed for training. In my case I only have 5-6 labeled data points. Rest 600 data points are non labeled.Is a semi-supervised approach still viable? Or should another approach be followed?
I thought about preprocessing the data and applying the unsupervised anomaly detection techniques but that changed the predictions observed.