Cost-sensitive feature selection via the ℓ2,1-norm

摘要: Abstract An essential step in data mining and machine learning is selecting a useful feature subset from the high-dimensional space. Many existing selection algorithms only consider precision, but do not error types test cost. In this paper, we use l 2 , 1 -norm to propose cost-sensitive embedded algorithm that minimizes total cost rather than maximizing accuracy. The with joint minimization of loss function misclassification costs. based costs robust outliers. We also add an orthogonal constraint term guarantee each selected independent. proposed simultaneously takes into account both Finally, iterative updating provided using objective makes more efficient. realistic algorithms. Extensive experimental results on publicly available datasets demonstrate effective, can select low-cost achieve better performance other real-world applications.