AbstractThis thesis is concerned with the use of Artificial Immune System (AIS)in the area of Ambient Assisted Living (AAL). The hypothesis for the work presented herein is that the AIS features of self-learning and adaptability address the complex problem of improving the detection of unknown abnormal; behaviour in the long-term monitoring of the elderly.
The work presents and affordable Open Hardware Data Acquisition Device that in combination with a Markov chain-based software simulation environment can be used for the collection of human activity data and the generation of necessary information for long-term simulation. The main contributions from the work presented herein relate to the design and use of AIS based solutions, and the selection of appropriate parameter combinations for supervised classifiers. Firstly, a novel seeding technique for AIS is presented that improves the placement of detectors in the search space. Secondly, a novel AIS-based monitoring algorithm, inspired by Hierarchical Temporal Memory architecture, is designed to learn and approximate sensor data to detect and report activity abnormalities. Thirdly, an empirical analysis is carried out to provide a clear understanding of how sampling frequency, segmentation method, window size, and computational load in an area of AAL. Fourthly, a Pareto curve based technique has been devised and demonstrated as a useful tool for the informed selection of parameter combinations to achieve the best possible classification accuracy and computational load.
The evaluation of the AIS-based algorithm showed that the detection rate of abnormal activity outperformed the results of supervised classifiers with parameter combinations selected based on the Pareto curve. The results are encouraging and support the decision to introduce the use of AIS for the detection of abnormal activity in AAL environments.
|Date of Award
|Djamel Azzi (Supervisor), Rinat Khusainov (Supervisor) & Branislav Vuksanovic (Supervisor)