Improved personalised data modelling using parameter independent fuzzy weighted k-nearest neighbour for spatio/spectro-temporal data
Machine learning technologies have been growing rapidly in recent years. Researchers have come up with several data processing architectures, enabling machines to consume, interpret, and produce understandable output from real-world data to improve the quality of our lives. The NeuCube architecture...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/1064/1/24p%20MOHD%20HAFIZUL%20AFIFI%20BIN%20ABDULLAH.pdf http://eprints.uthm.edu.my/1064/2/MOHD%20HAFIZUL%20AFIFI%20BIN%20ABDULLAH%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/1064/3/MOHD%20HAFIZUL%20AFIFI%20BIN%20ABDULLAH%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Machine learning technologies have been growing rapidly in recent years. Researchers have come up with several data processing architectures, enabling machines to consume, interpret, and produce understandable output from real-world data to improve the quality of our lives. The NeuCube architecture is a data processing architecture for spatio/spectro-temporal data which consists of four main modules: a spike encoding module, a recurrent SNN reservoir, an output module, and an optimization module. Despite it has been utilised on many various applications, most improvement of the architecture focuses on user experience rather than improving the result accuracy. Upon exploration of the architecture, the weighted k-nearest neighbours algorithm used for the classification module is found to be prone to misclassification as it relies solely on the majority voting rule to determine the class for new data vector. Additionally, it does not consider the class-specific fuzzy weight information during the classification process. Therefore, a data modelling mechanism which implements PIfwkNN classifier algorithm for improving the overall classification accuracy of the NeuCube architecture has been proposed. The proposed data modelling applies an additional class-specific fuzzy weight information to new data vectors during the classification process. In this research, the optimal parameters set for experiments has also been identified. The approach has been validated by using the Kuala Krai Rainfall Dataset, Dow Jones Index Data Set, and Gold Price and Performance Dataset for the 3-days earlier and 1-day earlier event prediction. From the experiments, the improved personalised data modelling using PIfwkNN classifier has shown a significant increase in terms of overall classification accuracy as compared to the conventional MLP, fkNN, and NeuCube with wkNN classifier. |
|---|