IMPACT OF NUMBER OF ATTRIBUTES ON THE ACCURACY OF HUMAN MOTION CLASSIFICATION

Abstract

The quality of the human motion data faces challenges in producing high classification accuracy in large data streams for essential knowledge discovery. This reflects the need to identify the key factors that affect the results of classification. Present studies merely focus on estimating joints, skeleton and motions of human activities. However, the effect of the number of attributes towards classification accuracies of human motion has not been discussed. Therefore, this paper is aimed at determining the amount of attributes that affect the qualities of human motion classification. The case studies involve simple locomotion activities: jumping, walking and running retrieved from the public available domain. The raw video data were transformed into numeric in the form of x and y-coordinates and rotation angles as to be tested from a single up to triple combinations of data attributes. The impact of the number of attributes on classification accuracy is evaluated via Bayes, Function, Lazy, Meta, Rule and Trees classifier algorithms supported by the WEKA tool. Results revealed that three attributes data gave the best classification performance with an average accuracy of 81.50%. The findings also revealed that the number of attribute is directly proportional to the classification accuracy of human motion data.

Keywords

• Attributes
• Classification
• Human motion
• WEKA