IncMSTS-PP: An Algorithm to the Incremental Extraction of Significant Sequences from Environmental Sensor Data

Abstract

The mining of sequential patterns from environmental sensor data is a challenging task. The data can present noises and contain sparse patterns hide in a huge amount of information. The knowledge extracted from environmental sensor data can have many applications: indicate climate changes, risk of ecologic catastrophes and help to determine environment degradation in face of humans actions. However, there is a lack of methods that can handle this kind of data. Based on that, we proposed IncMSTS-PP: an incremental algorithm that finds sequential sparse patterns and enhances them semantically facilitating the interpretation. IncMSTS-PP implements STW-method to extract stretchy patterns (patterns with time gaps) in data with noises. The enhancement use post-processing method that generalizes the patterns using the fuzzy ontology knowledge. Our experiment shows that IncMSTS-PP extracts 2.3 times more relevant sequences than traditional algorithms in sensor domain. The post-processing summarizes the patterns reducing to 22.47% of the original number of patterns. In conclusion, IncMSTS-PP is efficient and reliable in the extraction of significant sequences from sensor data.