Daily Activity Feature Selection in Smart Homes Based on Pearson Correlation Coefficient

• Published in: Neural processing letters Vol. 51; no. 2; pp. 1771 - 1787
• Main Authors: Liu, Yaqing, Mu, Yong, Chen, Keyu, Li, Yiming, Guo, Jinghuan
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2020; Springer Nature B.V
• Subjects: Activity recognition; Artificial Intelligence; Complex Systems; Computational Intelligence; Computer Science; Correlation coefficients; Datasets; Machine learning; Neural networks; Ontology; Sensors; Smart buildings; Smart houses; Activity recognition; Pearson Correlation Coefficient; Feature selection; Smart home
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-019-10185-8

In the case of a smart home, the ability to recognize daily activities depends primarily on the strategy used for selecting the appropriate features related to these activities. To achieve the goal, this paper presents a daily activity feature selection strategy based on the Pearson Correlation Coefficient. Firstly, a daily activity feature is viewed as a vector in Pearson Correlation Coefficient formula. Secondly, the relation degree between daily activity features is obtained according to weighted Pearson Correlation Coefficient formula. At last, redundant features are removed by the relation degree between daily activity features. Two distinct datasets are adopted to mitigate the effects of the coupling of the dataset used and the sensor configuration. Three different machine learning techniques are employed to evaluate the performance of the proposed approach in activity recognition. The experiment results show that the proposed approach yields higher recognition rates and achieves average improvement F-measures of 1.56% and 2.7%, respectively.