PT Unknown
AU Pierluigi Casale
TI Approximate Ensemble Methods for Physical Activity Recognition Applications
PY 2011
AB The main interest of this thesis focuses on computational methodologies able toreduce the degree of complexity of learning algorithms and its application to physicalactivity recognition.Random Projections will be used to reduce the computational complexity in Multiple Classifier Systems. A new boosting algorithm and a new one-class classificationmethodology have been developed. In both cases, random projections are used forreducing the dimensionality of the problem and for generating diversity, exploiting inthis way the benefits that ensembles of classifiers provide in terms of performancesand stability. Moreover, the new one-class classification methodology, based on an ensemble strategy able to approximate a multidimensional convex-hull, has been provedto over-perform state-of-the-art one-class classification methodologies.The practical focus of the thesis is towards Physical Activity Recognition. A newhardware platform for wearable computing application has been developed and usedfor collecting data of activities of daily living allowing to study the optimal featuresset able to successful classify activities.Based on the classification methodologies developed and the study conducted onphysical activity classification, a machine learning architecture capable to provide acontinuous authentication mechanism for mobile-devices users has been worked out,as last part of the thesis. The system, based on a personalized classifier, states onthe analysis of the characteristic gait patterns typical of each individual ensuring anunobtrusive and continuous authentication mechanism
ER