I have a time series data, which contains information from various sensors measured at every 20 minutes interval. I would like to use information from all these sensors as features to a Deep Learning/Machine Learning model (anything which is state of art at present for unsupervised anomaly detection would be great). There are no labels of anomalies in the data, and the model would need to be completely unsupervised to identify which groups of data fall into outliers, while which are normal behaviour.
In the same aspect, I would like to know the key features for each time-step, which are being used by the model to identify the behaviour as outlier/normal. So, for every time-step/row in my data, is there a good implementation in Python which can help me identify outliers in a fully unsupervised manner, with the cause too for each time-step/row?