It should be unsupervised feature learning. Unsupervised learning can show you cluster and pick features. I think you should first consider these backgrounds.

Take Autoencoder, learn the features in unsupervised manner and then pass to any classifier.

OR else,

" the famous object detection system called Deformable Parts Model basically 'learns' parts (the latent variables in the so-called Latent SVM formulation) in a straightforward iterative EM loop. Basically for learning, it initializes N parts as (smaller) bounding boxes located at fixed locations within and relative to the object bounding box; and then iteratively trains and validates part-level HoG detectors (SVM classifiers trained on HoG encoded image patches) moving around the part bounding boxes a little bit to get parts that give the best object-level detection performance! ( Zeeshan Zia )"

Dear Md Junayed Hasan,

Thank you for your time and answer..

I will go through it...

In line with MD Junayed,

Unsupervised learning used for clustering which means the structure by cluster could be derived of the data based on relationships among the variables in the data.

With unsupervised learning there is no feedback based on the prediction results.

I think its better to first learn the feature and then classify it through unsupervised learning. or if you already know the classification of features use supervised learning in second stage.