Can we apply the same algorithm to symbolic features and texture features?
Like for classifiers, it is basically impossible to say which is the best algorithm for feature selection or extraction.
It all depends on your application at hand. Few things you should keep in mind are:
- What kind of problem are you trying to solve? e.g. classification, regression, clustering, etc.
- Do you have a lot of data?
- Do your data have very high dimensionality?
- Is your data labelled?
Once you have answers to these questions you can decide:
- Do you want to perform feature extraction or feature selection?
- Do you want to use a supervised or unsupervised method?
- Do you want to use a very computationally intensive method or something rather inxepensive?
If you tell a bit more about your problem I can try to give you some hints!
Marco
Like for classifiers, it is basically impossible to say which is the best algorithm for feature selection or extraction.
It all depends on your application at hand. Few things you should keep in mind are:
- What kind of problem are you trying to solve? e.g. classification, regression, clustering, etc.
- Do you have a lot of data?
- Do your data have very high dimensionality?
- Is your data labelled?
Once you have answers to these questions you can decide:
- Do you want to perform feature extraction or feature selection?
- Do you want to use a supervised or unsupervised method?
- Do you want to use a very computationally intensive method or something rather inxepensive?
If you tell a bit more about your problem I can try to give you some hints!
Marco
Ultra sound image classification
i want to perform texture feature extraction or feature selection?
here i am enclosing some papers these will apply to texture images
well, i guess if your data is images, you would rather do feature extraction and then, if necessary (the vector of extracted features is still large and probably redundant), you can perform feature selection.
Some very popular features include for example SIFT or SURF, but there are many other features. I think you have to perform this choice based on what are the salient characteristics that will allow you to differentiate the classes of textures you have in mind.
I am sorry but i am not sure i got it right. Please correct me if I am wrong.
Do you extract 5 features from every image and then you classify them based on this 5 features?
If so, your feature space has already low dimensionality, so I guess you can just train a classifier on these 5 features and see whether it works well.
Another thing you can do (given that you have only five features) is to exhaustively search all the possible features sets which are only 32 (2^5), and therefore the search should be not too time consuming.
You can also compute co-occurrence distributions (e.g. GLCM) on the texture and get more features just to increase a dimensionality of your feature vectors. Then you can use WEKA data mining tool to find the most important features. I do not know if it is applicable in your case, but generally if you have several image representing the same object, you can apply PCA on these images and use the most relevant principal component as “selected features”.
You could try PLOFS: It is a fast wrapper approach with a subset evaluation satisfying monotonicity.
Jiang Li, Jianhua Yao, Ronald M. Summers, Nicholas Petrick, Michael T. Manry, and Amy K. Hara, “An Efficient Feature Selection Algorithm for Computer-Aided Polyp Detection,” special issue of the International Journal on Artificial Intelligence Tools (IJAIT), vol. 15, no. 6, December 2006, pp. 893-915.
Jiang Li, Michael T. Manry, Pramod Narasimha, and Changhua Yu, “Feature Selection Using a Piecewise Linear Network”, IEEE Trans. on Neural Networks, Vol. 17, no. 5, September 2006, pp. 1101-1115.
Feature extraction is highly subjective in nature, it all depends on what type of problem you are trying to handle. There is no generic feature extraction scheme which works in all cases.
If you are handling images, you extract features (appropriate) and if the feature dimension is high then try to do the feature selection or feature transformation using PCA where you will get high quality discriminant features.
You can go for Evolutionary algorithms they yield more practical results
Its difficult to say which feature extraction and selection method is good. It is highly subjective in nature and very specific for application at hand. Coming back to your question regarding symbolic data handling. May not be true always.
Before, feature extraction or feature selection, feature definition is an important step, and actually it determines the core of the solution. If you are starting from the point after that step, i.e., the features are pre-fixed, the Marco's comments above is a good starting point.
Principal Component Analysis is one of the most popular technique. PCA is a feature extraction method. You can perform it in matlab easily.
Feature extraction and feature selection are two techniques tied to hand crafted features. From my experience, if you have a wide matrix (more features than data points), lasso/lars might be a good choice. If you have a tall matrix (more data points than features), on the other hand, the PLOFS algorithm mentioned above might be used. However, if you have a large date set for this case, you might try deep learning on raw data to automatically extract features and try to replace the feature engineering.
Please look into it
http://iajit.org/PDF/vol.10,no.6/4134.pdf
Article Content Based Image Retrieval Using Combined Features of Sha...
Hi
I am a new in ML and I would like create dataset like CIFAR10 and I couldn't find any clear exmples
With the increasing research with deep learning, I could suggest try with convolution neural networks and you don't need to be worried for features or how to select them.
Thanks for your answer and I have my own dataset and I would to create them from the scratch. Do you have any idea about how I can create these data in order to train my Machine Learning
First, I suggest you have images with well separated features after you can use cross validation.
you mean for example I should have two sepeated files
one : it contains a set of images with labels to train ML
second: it contains a set of images with labels to test ML.
So my question is how I can map them together my images
You can use SURF algorithm (Speeded Up Robust Features) for
feature extraction and feature selection. The main advantages of using this algorithm are this is Rotational invariant and faster in computation because here we deal only 64 Dimension vectors.
better to use PLOFS: It is a fast wrapper approach with a subset evaluation satisfying monotonicity.
For feature extraction and selection, decision tree, naivebays classification required
PCA is generally only good if the feature space is not terribly large, because the computational efficiency is very low. It will work with other feature extraction methods to further define the features that are best for your particular problem, though. However, if you want a faster way, you can use a classification method that uses an L1-norm function. This will naturally reduce some coefficients to 0 and they will effectively become drop out terms. However, with a data set that has many parameters this still might not reduce the parameters enough for your particular problem.
There are many feature selection techniques that are frequently used now in these days, such as Independent Component Analysis(ICA), Principal Component Analysis(PCA), and Linear Discriminant Analysis(LDA). You can perform it in Matlab
The best feature textural extractors are local Binary Patterns (LBP) and BSIF.
The best feature extraction algorithm depends on the application
The Algorithm depends on the application. There are a lot of them.
completing the answer, you can use: Independent component analysis
Isomap
Kernel PCA
Latent semantic analysis
Partial least squares
Principal component analysis
Multifactor dimensionality reduction
Nonlinear dimensionality reduction
Multilinear Principal Component Analysis
Multilinear subspace learning
Semidefinite embedding
Autoencoder
It depends on your application, you can use texture features, HOG, color features and histograms, PCA and so on.
If you want to learn the manifold, then local linear embedding is also a good option to consider.
Machine learning algorithms such as random forest and support vector machine can be applied to Varieties of data sets .
For image feature extraction kernel PCA method ,partial least squares, etc.
This algorithm depends on the problem and application.We can use different Extraction like Correlation Matrix with Heatmap .
Random Forests are often used for feature selection in a data science workflow.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification