This paper presents a generalization of the bagging procedure by using smoothed bootstrap with bagging, a procedure we call generalized bagging. Our generalized bagging method unifies input and output smearing, in the sense that noise is added to both the input and the output, so that input smearing and output smearing become special cases of generalized bagging. We discuss the choice of optimal smoothing parameter to control the variance of the added noise in the smoothed bootstrap. Our simulation studies show that the proposed method outperforms other competing methods, when the variance of the error term is large. We also demonstrate the performance of the proposed procedure with real datasets.
All Science Journal Classification (ASJC) codes
- Statistics and Probability
- Bandwidth selection
- Density estimation
- Smoothed bootstrap