This chapter introduces a hybrid algorithm for classifyingboth general human behavioral patterns as well as specific actions. Behaviormodeling is a significant research area in this field and has beenan active research topic for several applications. Achieving a high accuracyclassification model has also been a research target which derivedmany efforts for improving classification performance. Tuning classifictionparameters and data features are vital methods for obtaining accurateclassifier in any application. This work focuses on tuning classificationparameters of SVM classifier and optimizing features set usingElephant Herding Optimization Algorithm (EHO). The proposed classificationmodel will be validated using behavioral data obtained from amotion capturing camera system known as VICON to accurately classifynormal and aggressive behavioral patterns and specific actions.
Keywords: Behavior, Classification, Optimization, Vicon, Elephant Herding (EHO), Support Vector Machines (SVMs).1 INTRODUCTIONOptimizing classification parameters and classification features is a rich researchtopic. In this chapter meta-heuristic swarm-based algorithm is devoted to finetuneclassification parameters to get best performance for human behavior identification.Numerous applications of our daily life involve machine learning procedures.Currently, there are many sources that are rich of behavioral informationsich as images, video records, and many types of sensors where the internet ofthings comes (IOT) 1,2,3.
Such data is analyzed and introduced to aid real lifeapplications such as security, entertainment and medical diagnosis4. Lately,surveillance systems are surrounding us everywhere in the street, markets, andorganizations from simplest to most complicated ones 5,6. Modern systems aresupported with detection techniques for dubious behavioral patterns, classifyingpeople motions into normal and abnormal patterns to respond properly and inreal time 7.Many research efforts have been applying machine learning techniques (MLs)for the purpose of human actions identification.
Support Vector Machines (SVMs),Neural networks and Support Vector Regression (SVR) are examples 8. Theproper adjustment of learning parameters is a problem facing model designerswho sake for acheiving a model that accomplishes high generalization as well ashigh classification accurac 9. Other main problem faceing ML model designersis feature selection. Feature selection is concerned with filtering the input set offeatures to only most discriminant and most features relevant to the data. In thischapter, we discuss how Swarm Intelligence (SI) had a great share of machinelearning optimization research in past years 10,11.
In this work we show howSI could be employed to solve the problem of classification parameters tuning aswell as features selection using selected SI algorithms.