Methodology for improving classification accuracy using ontologies: application in the recognition of activities of daily living

Abstract

Feature construction and selection are two key factors in the field of machine learning (ML). Usually, these are very timeconsuming and complex tasks because the features have to be manually crafted. The features are aggregated, combined or split to create features from raw data. In this paper, we propose a methodology that makes use of ontologies to automatically generate features for the ML algorithms. The features are generated by combining the concepts and relationships that are already in the knowledge base, expressed in form of ontology. The proposed methodology has been evaluated with three different activities of a popular dataset, showing its effectiveness in the recognition of activities of daily living (ADL). The obtained successful results indicate that the use of extended feature vectors provided by the use of ontologies offers a better accuracy, regarding the original feature vectors of the classic approach, where each feature corresponds to the activation of a sensor. Although the classic approach produces classifiers with accuracies above 92%, the proposed methodology improves those results by 1.9%, on average, without adding more information to the dataset.

Ontology-Based Framework for the Automatic Recognition of Activities of Daily Living Using Class Expression Learning Techniques

“…e framework proposed in Section 4 has been used to convert both datasets to all the available ontology models for the description of ADL, namely, Salguero [3], SPHERE [21], COSAR [22] and Noor [23]. In addition, a version with the OMA ontology has been generated in which information about the context has been included, following the scheme proposed in Section 4.3.…”

“…In the literature, there are many proposals that employ ontologies for the recognition of ADL. In previous works, we also proposed an ontology for the representation of ADL [3,8]. However, this ontology was focused solely on the identification of the events produced by the sensors, ignoring the rest of the information about the activities, such as the types of the sensors involved or the rooms where they were located.…”

“…e events are only related to the time interval in which they occur, but there is no order relationship among the events in the interval. e ontology we have proposed in previous works [3] employed the opposite approach, where the description of activities as sequences of events produced by the sensors were considered, ignoring the contextual information. For this reason, we propose in this work a generic ontology that allows the description of ADL as sequences of sensors events as well as including other kinds of contextual information of activities, such as their location or the posture of the inhabitants.…”

“…In this field, many proposals have been presented in the literature, many of them being based on logical rule systems. Among them, ontology-based approaches have provided very successful results in both data-driven approaches (DDAs) [3,4] and knowledge-driven approaches (KDAs) [5][6][7]. DDA usually relies on machine learning techniques in which a preexistent annotated dataset of user behaviors is required.…”

“…However, none of the available ontologies in the literature considers the order in which events occur. Aware of this situation, we proposed an ontology in a previous work specifically designed to represent the sequences of sensor events [3,8]. It was, actually, the only information about activities considered in the ontology.…”