Please use this identifier to cite or link to this item:
|Title:||Cognitive systems for data analytics||Authors:||Kwan, Sok Wei||Keywords:||DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence||Issue Date:||2016||Abstract:||Currently, most of the research focuses on recognition and analysis of low level activities of human such as walking or sitting instead of analysis of high level activities such as human activity pattern. Hence, this project is mainly targeted at studying how cognitive systems may be used to allow the autonomous discovery of hidden human pattern from datasets and find behaviors of groups of people who have common characteristics. The scope of the project would be to produce demographic group models through analysis of data sets using cognitive-related algorithms. Datasets that show detailed activity patterns of respondents were reviewed for analysis. There was a round of data pre-processing to remove any unwanted or irrelevant variables. After that, the domains of activity patterns to be explored were determined and relevant variables were selected for the preparation of datasets. Algorithms to be used for analysis were determined after research and used to analyze the data sets prepared. Adaptive Resonance Theory (ART)-based algorithms such as Fusion ART, Fusion ART-with Match Tracking, Fusion ART-with Interesting Feature and IFC ART were identified to analyze the data sets. They are used instead of the more commonly known clustering techniques such as K-means or X-Means for a few reasons. ART based algorithms aforementioned have been shown to have low computational complexity with the ability to adaptively learn, and more importantly they are cognitive. The best performing algorithm among the four ART-based algorithms, according to Davies-Bouldin Index (DBI), would be determined for each data set and used to run the analysis to derive the demographic group models. There are seven case studies done in this project. Some interesting findings have been uncovered from the demographic group models in the case studies such as the industries that majority of low earners work in, industries that high earners work in and common characteristics of those who call at odd times of the day. Anomalies could be found, and one that was found is a 15 years old male is a high earner by working with Construction and Extraction jobs in Construction industry. Limitations posed in this project are that the results are only representative of Americans and also the results only target specific few activity patterns. Future work or research could be done by applying the algorithms on data sets that is representative of the population in other countries or other activity patterns such as leisure activities in ATUS data set.||URI:||http://hdl.handle.net/10356/66604||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Student Reports (FYP/IA/PA/PI)|
Updated on Jun 22, 2021
Updated on Jun 22, 2021
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.