Statistical features gives a probability one for all features, while importing ensemble table to diagnostic feature designer.

Question

sed 2022-10-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1832928-statistical-features-gives-a-probability-one-for-all-features-while-importing-ensemble-table-to-di

回答： Akshat Dalal 2024-1-1

I want to select statistical features from raw signals data obtained for four different faultCodition, yet when uploading the time table shown below to the Diagnostic Features Designer App. All the statistical features appear to have a probability of one for different faultCodes. Kindly any hint on this to rank the features importance.

2 个评论
显示无隐藏无

Image Analyst 2022-10-22

What toolbox is this in? And you forgot to attach your data.

sed 2022-10-22

I am importing the data shown in the screenshots to Diagnostic features designer app.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Akshat Dalal 2024-1-1

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1832928-statistical-features-gives-a-probability-one-for-all-features-while-importing-ensemble-table-to-di#answer_1381101

Hi Sed,

I understand that you want to rank features using the Diagnostic Feature Designer App but are getting a probability of one for different fault conditions. This might indicate that the features are not discriminative enough for the fault conditions you are trying to classify. To address this issue and rank the feature importance more effectively, you can try the following approaches:

Feature Engineering: Experiment with different types of features, including time-domain, frequency-domain, and time-frequency domain features. Features such as FFT coefficients, wavelet transforms, or higher-order statistics might provide better discrimination. To read more, please refer the following documentation: https://www.mathworks.com/discovery/feature-engineering.html
Class Balancing: If your dataset is imbalanced, use techniques such as oversampling the minority class, undersampling the majority class, or generating synthetic samples with SMOTE to balance the classes before feature selection. To read more, please refer the following documentation: https://www.mathworks.com/help/stats/classification-with-imbalanced-data.html
Cross-validation: Use cross-validation to evaluate the discriminative power of features across multiple subsets of your data to ensure that the feature importance ranking is robust. To read more, please refer the following documentation: https://www.mathworks.com/discovery/cross-validation.html
Visualization: Use visualization tools such as t-SNE or UMAP to visualize the separability of the classes in the feature space. This can give you insights into whether the features are capable of distinguishing between different fault conditions. To read more, please refer the following documentation: https://www.mathworks.com/help/stats/t-sne.html