摘要
FDR(Fractal Dimensionality Reduction)算法的主要问题在于需要多次扫描数据集,I/O开销比较大,Opt-FDR(Optimized FDR)通过对FD-tree进行动态调整来避免多次扫描数据集,但对算法的空间需求比较高.借鉴Z-ordering索引技术的思想,设计并实现了一种改进的分形属性选择方法ZB-FDR(Z-ordering Based FDR).该方法仅需要扫描数据集一遍建立底层网格结构,基于该底层网格结构实现分形维数的计算及后向删除维操作.在合成数据集及实际数据集上的实验结果表明ZBFDR具有较为优良的整体性能.
FDR （Fractal Dimensionality Reduction） is the most famous fractal dimension based feature selection algorithm proposed by Traina in 2000. However,it is inefficient in the high dimensional data space for multiple scanning the data set. The optimized algorithm proposed in 2004, Opt-FDR, which scans the data set only once and adjusts the FD-tree dynamically for evaluating the fractal dimension. But the adjusting process is complicated and the space complexity is high. Taking advantage of the Z-ordering technique, we propose an optimized FDR, ZB-FDR（Z-ordering Based FDR）,which scans the data set only once to initialize the lowest cell queue and performs the backward feature reduction. The experimental results show that ZB-FDR algorithm achieves better performance.
出处
《兰州交通大学学报》
CAS
2007年第1期6-10,25,共6页
Journal of Lanzhou Jiaotong University
基金
光电技术与智能控制教育部重点实验室(兰州交通大学)开放基金资助项目(K04116)
甘肃省教育厅科研项目资助(0604-09)
