Knowledge-Driven Complex Human Activity Recognition using Ontologies: A Case for Muslim Prayer (Salah)

Abstract In this data-driven landscape, sensor-based activity recognition has transformed our understanding of human behaviour by enabling knowledge extraction from diverse sources. This study focuses on Mus- lim prayer (Salah), a ritual observed five times daily by over 1.8 billion Muslims worldwide. By merging knowledge-driven methodologies with sensor-based activity recognition, this research augments the standardised modelling of Salah through the ”SemanticSalah ontology”. This formal knowledge model encapsulates all facets of Salah execution, including its sequential progression, evaluated through rule languages such as Description Logic and Semantic Web Rule Language, enabling critical inferences about Salah. The primary objective is to empower practitioners with tools for precise and complete Salah observance. The SemanticSalah ontology is evaluated through a hybrid approach combining criteria-driven, expert, and data-driven methods. Expert evaluations gauge the competency of the ontology. Data collected from 49 prayer units, captured through smartwatches worn by different users, is utilised for human activity recognition, enriching the knowledge graph that covers diverse Salah scenarios and fortifying reasoning and inference mechanisms. The SemanticSalah ontology and knowledge graph are freely accessible at https://github.com/A-Kamran/SemanticSalah.