A Two-stage Gait Analysis Using Interval-based Sequential Pattern Mining

Abstract

Artificial Intelligence (AI) has recently experienced a significant rise in its implementation, as evidenced in several industries, such as telerehabilitation. Regarding the digital transformation in telerehabilitation, some companies have already utilized machine learning techniques to analyze the dataset collected via sensors or cameras and then support distance medical experts in diagnosing through the Internet. This research proposes a novel two-stage data mining framework combining gait analysis and interval-based sequential pattern mining. Potential problems or diseases are discerned in the first stage by analyzing gait videos captured from a camera. A series of walking postures captured frame by frame can be transposed into a sequence of events. For instance, a particular frame might depict a gait indicative of a possible Parkinsonian gait. Since these events are recorded temporally, a series of similar events can be merged to form an interval-based event, described by its starting and ending points. Subsequently, the second stage involves extracting and recognizing patterns from a dataset of interval-based temporal sequences through sequential pattern mining. This preliminary experiment collected twenty real-life samples and corroborated the usefulness of the proposed model.