Over the last decade, sensory data has had a massive impact on all aspects of our daily lives. Smartphones, in addition to wearables (e.g., smartwatch, smartglasses, etc.), provide continuous measurements over time captured from different attached sensors (e.g. GPS, Bluetooth, compass, accelerometer and proximity sensor). This one-dimensional data is fraught with valuable information that can express different patterns. Consequently, several approaches have been developed to extract features and classify these sequential measurements into understandable classes.
Latent Dirichlet Allocation (LDA) is a well-known approach in topic modelling that is basically used to cluster documents into hidden topics based on their observed contents. By considering sensory data as a set of words, LDA is supposed to be capable to cluster this data and thus understand and extract unobserved patterns. One of the important patterns is detecting anomalies (outliers) from sensory data. Anomalous patterns can be defined as the patterns which occur rarely compared to normal ones. For example, detecting suspicious displacement of the smartphone, given the measurements of the GPS, the accelerometer and/or gyroscope, can help in several security issues. Another example is automatically detecting Arrhythmias (Heart Rhythm Disorders), given the measurements of the Heart rate monitor, which is a serious problem for elderly people.