Self-calibrated approach for fisher discriminant classifier estimation in linear discriminant analysis using linear programming
As far as what we have commonly consider theoretically, data sets always have enough samples and limited features to analyze. For such data, parameters can be substantially refined as the sample size increases toward infinity. However, the real-world data are more complicated and limited, especial...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/148491 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | As far as what we have commonly consider theoretically, data sets always have enough samples and limited features to analyze. For such data, parameters can be substantially refined as the sample size increases toward infinity.
However, the real-world data are more complicated and limited, especially in industries such as Finance, Biology and Medical Science. The data under analysis are high dimensional and sparse, meaning that there may be far more features than the size of the given data set. Sparse terms and singular matrices all add difficulty to parameter estimation.
In order to solve the problem with the sparse term, this paper introduces a self-calibrated estimator to estimate the Fisher’s linear discriminant classifier that is tuning-insensitive. The new method does not require cross validation over parameters, thus, enjoys better timing performance and rate of convergence theoretically. We further demonstrate the performance of the proposed method through numerical simulations. |
|---|