Abstract: This paper presents a new method that combines deep k-means clustering with granule mining approaches to utilise contextual information for improving outlier detection and classification.