Sparse kernel learning-based feature selection for anomaly detection

PROJECT TITLE :

Sparse kernel learning-based feature selection for anomaly detection

ABSTRACT:

In this paper, a completely unique framework of sparse kernel learning for support vector knowledge description (SVDD) primarily based anomaly detection is presented. By introducing 0-one management variables to original features within the input house, sparse feature choice for anomaly detection is modeled as a mixed integer programming downside. Due to the prohibitively high computational complexity, it's relaxed into a quadratically constrained linear programming (QCLP) problem. The QCLP drawback will then be practically solved by using an iterative optimization methodology, in which multiple subsets of options are iteratively found versus a single subset. But, when a nonlinear kernel like Gaussian radial basis operate kernel, associated with an infinite-dimensional reproducing kernel Hilbert area (RKHS) is utilized in the QCLP-primarily based iterative optimization, it is impractical to seek out optimal subsets of features thanks to a giant range of doable combinations of the first options. To tackle this issue, a feature map called the empirical kernel map, which maps data points in the input space into a finite house referred to as the empirical kernel feature house (EKFS), is utilized in the proposed work. The QCLP-based iterative optimization downside is solved in the EKFS rather than within the input space or the RKHS. This is often attainable as a result of the geometrical properties of the EKFS and the corresponding RKHS remain the same. Currently, an specific nonlinear exploitation of the information during a finite EKFS is achievable, which ends up in optimal feature ranking. Comprehensive experimental results on 3 hyperspectral pictures and many Machine Learning datasets show that our proposed technique will give improved performance over the current state-of-the-art techniques.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here