Converted both the files Into statistical file to perform the analysis. Opened the training statistical file, attached the Type node, and selected the target. Attached Neural network node to the Type node, selected the basics option under the build option tab and by selecting the customize number of units, assigned 3 In the hidden layers . Run the Neural network analysis and attached the Analysis node to the golden nugget to check the results. Followed the last two steps and created two different neural network models by assigning 5, 9 and 15 in hidden layer respectively.
Stopping rules set to 15 minutes per model, used default values in ensembles, overbite prevention set to 30%. Lastly, attached the Analysis node to each of the golden nugget retrieved, to check the results for the neural network Model 1: Hidden layer – 3 (results) Model 2: Hidden layer – 5 (results) Model 3: Hidden layer – 9 (results) After getting the above results, I loaded the testing statistical file, attached the Type node and selected the target as bankruptcy and attached all the four golden nuggets to it. By using the analysis node, below are the results for my neural networks. Stratified Sampling
I have create two different files, one with all the Xi’s (yes) and other will all the Co’s (no) using excel. File conversion using SPAS statistical, converted the excel files to statistical files (. Save). Loaded the statistical files in SPAS modeled, using the partition node, created two different files having 25 values of Xi’s each, similarly did the same for the files having all Co’s. Exported using excel node from the output tab. After that, using excel, created two different files by picking 25 values form each earlier created files in order to make testing and training data sets for the analysis.
Loaded the training file in SPAS modeled, and performed neural network modeling with same hidden layers 3, 5, 9 and 15. Stopping rules set to 15 minutes per model, used default values in ensembles, overbite prevention set to 30%. Attached the analysis node to each of the golden nugget retrieved for final analysis. Mode: Hidden layer – 15 (results) After getting the results for each neural network loaded the testing file and attached the type node and selected bankruptcy as target. Attached all the four golden nuggets retrieved during above analysis, and attached the analysis node to see the exults.
Confusion Matrix Training data Testing data Correct Wrong Model 1 74% 36% 26% Model 2 76% 24% Model 3 32% Model 4 34% 66. 67% 33. 33% 68. 89% 31. 11% 37. 78% 72% 28% Based on the finding and comparing the results attained via using both the sampling technique, I will choose Model 4 of random sampling. Firstly, represents the highest percentage for predicting the bankruptcy correctly and consequently, the lowest percentage for predicting wrong. Secondly, rate of improvement is the highest among other models used with different hidden layers.