哎,一片论文的实现,可是我还不了解这个,没有用软件使用过。
1. Collect a data set S in October which describes the input-
output relationship for DMUs. Obtain the preprocessed
data set SS after the data in S are divided by an integer
1000 for the purpose of neural network training and
simulation.
2. CCR method is used to calculate efficiency score of
DMUs in S. The data set SS is grouped into four
categories S1, S2, S3, and S4 based on the efficiency
scores. The efficiency score intervals of S12(0.98, 1] are
referred as ‘strong relative efficient interval’. The
efficiency score intervals of S22(0.8, 0.98] are referred
as ‘relative efficient interval’. The efficiency score
intervals of S32(0.5, 0.8] are referred as ‘relative
inefficient interval’ and the efficiencies of S42(0, 0.5]
are referred as ‘very inefficient interval’.
3. Train neural network NN1 with S1 and any other two
groups of data subset (e.g. S1g S2gS3 or S1gS2gS4
or S1gS3gS4). If the pre-specified epochs or accuracy
is satisfied, STOP; go to Step 5.Otherwise, change one
training subset and go to Step 3.
4. Apply the trained neural network model to the data set
SS to calculate efficiency scores of all DMUs.
5. Postprocess the calculated efficiency scores by regress
analysis between DEA-NN results and CCR DEA results.
Do statistics analysis of DEA-NN efficiencies.
6. Change the data set in October to those of November and
December. The above procedures (from Step 1 through
Step 5) are applied for each data set.
大家能帮我下么?
这个是先用dea求出有效和无效的决策单元,然后根据dea的score值进行分组。
再用bp神经网络训练。
麻烦各位了,如果你有同样地问题,
我愿意与你交流1019990976,paper的方向是分类,预测,具体topic是供应商选择问题。谢谢