search

nagdevAmruthnath / Predictive-Maintenance

A notebook tutorial series for performing predictive maintenance using machine learning
MIT License
144 stars 75 forks source link

Flawed analysis including incorrect predictive modelling for Predictive Maintenance - Machine Learning.ipynb #3

Open ghost opened 4 years ago

ghost commented 4 years ago

Note: Please, do not close the issue until its agreed by the OP as resolved! Its unethical

Feature extraction and feature selection are two different entities all together. They should not be confused to be the same.

Continuing further, the given answer makes no sense. Let me elaborate, Feature extraction does not mean to extract the feature summary statistic ! Feature Extraction means to build features based on existing feature-set. A famous example in unsupervised learning is principal component analysis (PCA). PCA is a linear transformation method that yields the directions (principal components) that maximize the variance of the data. The principal components is a good example of unsupervised feature extraction.

nagdevAmruthnath commented 4 years ago

It feels like OP created a new account just to comment here and to create an issue for this project.

Since thats out of the way.

In OP's own words feature extraction is build features based on existing feature data set. As explained in my previous post, Features extracted doesn't have to be from a model like PCA or autoencoders. It is very domain specific. Please review some signal processing and feature extraction techniques by Dr. Jay lee and other notable authors.

PCA is a linear transformation of features. Theya re most famously known for feature selection and dimensionality reduction.

Some reading material for OP. I will keep in open until OP is convinced. 🤣😂

https://www.sciencedirect.com/science/article/pii/S187770581403522X/pdf?md5=c292591d1df2281935a11a0626be217b&pid=1-s2.0-S187770581403522X-main.pdf

https://ieeexplore.ieee.org/iel7/6287639/6514899/08085101.pdf

https://www.academia.edu/download/39419294/10.0000_www.inderscienceonline.com_generic-9C3E1BD1473F.pdf

ghost commented 4 years ago

First, creating a new account or using an old account, is an inconclusive concern and there is no point grappling about it! I accidentally came across this flawed article! Not having a github account, I had to create one!

With that out of the way, shall we continue further!

Please, do not direct me to read literature, because there is no dearth of articles proffering that X is right and/or X is wrong. I too can give you references for the same. Here is some literature for the author to review wherein PCA is classified as either a feature extraction or a feature selection method,

https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8962035

https://www.sciencedirect.com/science/article/pii/S0959152419305062

Second, the author is still beating around the bush and not answering the initial question. The Q was the incorrect usage of summary statistic in cell 9 and subsequent modelling. I'm trying to engage the author in a constructive dialogue. For once from a discussion perspective forget about feature selection or extraction. Do you still not see the mistake in this analysis?