In this paper, we propose an approach to combine the kernel matrices constructed by sliced average variance estimation (SAVE) with various numbers of slices. The proposed approach is called fused sliced average variance estimation (FSAVE). By fusing the information by usual SAVE applications with different slice numbers, the sensitivity to slices can be reduced, so the structural dimension estimation can be improved. Numerical studies confirm this, and a real data analysis is presented.
- Inverse regression
- Sliced average variance estimation
- Sufficient dimension reduction