Multiclass Classification Method in Handheld Based Smartphone Gait Identification

Abstract

Gait identification has been widely used in many types of research and application. Since gait identification involves with many people and classes, using a single classifier is not a good option as the dataset may contains overlapped class boundary and moreover, most of the classifiers are well built for binary classes. This paper discusses the application of multiclass classifiers such as one-vs-all (OvA), one-vs-one (OvO) and random correction code (RCC) on handheld based smartphone gait signal for person identification. The mapping uses J48 as the main classifier. The result is then compared with a single J48 for the benchmark. Finally, the best multiclass method is compared with few machine learning classifier in-order to see its capability. From the result, it can be seen that using OvO and RCC thus increase the accuracy performance if compared to a single classifier. For the best classifier in the multiclass mapping method, it can be seen that J48 yield the best accuracy score.