Predictive Framework for Imbalance Dataset
The purpose of this research is to seek and propose a new predictive maintenance framework which can be used to generate a prediction model for deterioration of process materials. Real yield data which was obtained from Fuji Electric Malaysia has been used in this research. The existing data pre-pro...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | eng eng |
| Published: |
2012
|
| Subjects: | |
| Online Access: | https://etd.uum.edu.my/3603/1/s91447.pdf https://etd.uum.edu.my/3603/7/s91447.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The purpose of this research is to seek and propose a new predictive maintenance framework which can be used to generate a prediction model for deterioration of process materials. Real yield data which was obtained from Fuji Electric Malaysia has been used in this research. The existing data pre-processing and classification methodologies have been adapted in this research. Properties of the proposed framework include; developing an approach to correlate materials defects, developing an approach to represent data attributes features, analyzing various ratio and types of data re-sampling, analyzing the impact of data dimension reduction for various data size, and partitioning data size and algorithmic schemes against the prediction performance. Experimental results suggested that the class probability distribution function of a prediction model has to be closer to a training dataset; less skewed environment enable learning schemes to discover better function F in a bigger Fall space within a higher dimensional feature space, data sampling and partition size is appear to proportionally improve the precision and recall if class distribution ratios are balanced. A comparative study was also conducted and showed that the proposed approaches have performed better. This research was conducted based on limited number of datasets, test sets and variables. Thus, the obtained results are applicable only to the study domain with selected datasets. This research has introduced a new predictive maintenance framework which can be used in manufacturing industries to generate a prediction model based on the deterioration of process materials. Consequently, this may allow manufactures to conduct predictive maintenance not only for equipments but also process materials. The major contribution of this research is a step by step guideline which consists of methods/approaches in generating a prediction for process materials. |
|---|