Subscribe Bookmark RSS Feed

Please recommend best tool for correlating large number of input parameters to categorical outcome

novicescriptor

Community Trekker

Joined:

Oct 1, 2012

My question is best described using a sample table below:

Calibration ParametersPerformance Result
SNPAR1PAR2PAR3PAR4PAR5P/F
10353.4144.224.431158120.8P
11683.7546.694.601149121.6P
18343.9442.214.481144121.2P
20623.6344.764.441155121.2P
21923.6046.095.161150121.5F
26703.2446.783.891141121.3P
27123.7046.625.851143121.2F
32473.4748.265.571168121.0F
46793.6646.735.001142120.6F
47033.5546.074.601146120.8P
48953.4047.694.921149120.4P
50073.8246.684.621140121.3P
57873.3744.375.621150120.9F
59013.3644.614.241142120.8P
70903.7247.115.601161121.3F
80883.3048.256.991146120.3F
82593.3946.785.671156121.0F
83293.6145.234.951151121.6F
86743.3546.024.821144122.1F
90173.6448.245.671154121.3F

In the real case there are thousands of input adjustments or calibration data. The output is actually categorized into sections also, but if the high level  objective here is to take a black box approach to narrow down the field, so a more detailed analysis can be done. i.e, we want to know which parameters(s) rank high in affecting the outcome with high degree of correlation?  (pass if above/below a certain threshold, with increasing confidence as you move away from  the threshold).

I have successfully used the Partition tool to arrive at the answer using the data above - the answer is Parameter 3 at 4.95.

2671_pastedImage_0.png

My quandry is the large number of parameters in the thousands and ideally the tool can be scripted to 'float up' the key parameters only, with no interactive input from user.

I hope I have described the question adequately - feel free to ask if I have left out anything.

Thanks,

Barry