Biometrics-based user identification with optimal feature evaluation and selection
Recently, biometrics-based identification algorithms have gained popularity as a means of identifying a person using their unique behavioral characteristics such as gait or hand movement pattern. Classifications based on biometric features are broadly used in modern healthcare applications, including user identification, authentication, and tracking. The complexity and accuracy of classification algorithms largely depend on the size and the quality of the feature set used to build classifiers. In this thesis, we mostly focus on feature evaluation and selection as these are the essential steps to decide a small set of high-quality features to build accurate and efficient classifiers in user identification. We propose a novel and efficient approach to evaluate and select biometric features for user identification based on activity sensor data collected from the users’ wrists while they are walking. For each feature, we first generate an NRMSD matrix, each entry of which represents the similarity level of any two users. Based on the observations from NRMSD matrices, we define two heuristics, Farness Value and Farness Ratio to evaluate the quality of the feature. We evaluated a total of 72 features and selected 18 high-quality features based on our evaluation results. Finally, we train our data with different classifiers and select KNN as the best classification model. Compared to other feature evaluation and selection techniques, this approach is more efficient and yields a higher accuracy of 98.3%.