README.md

Neural Network

This part, contains project sample for Neural Network in Python.

Run Regression

In order to run the project please follow this link.

Results

In order to achieve a clustering method using Neural Network approaches, we have to come up with a Network design. In this problem, I have employed 5 different network structure, and then test via 5-Fold Cross validation. The result is depicted below as well as the network structures.

Model	Data	RSS	TSS	R^2	Precision	Recall	F-Score
model#1	Train Error	80.980	839.030	0.903	0.970	0.972	0.971
	Test Error	36.507	366.374	0.900	0.967	0.977	0.972
model#2	Train Error	110.708	839.030	0.868	0.939	0.974	0.955
	Test Error	56.772	366.374	0.845	0.928	0.971	0.947
model#3	Train Error	95.325	839.030	0.886	0.953	0.975	0.963
	Test Error	47.468	366.374	0.870	0.949	0.973	0.960
model#4	Train Error	79.120	839.030	0.906	0.968	0.970	0.969
	Test Error	37.372	366.374	0.898	0.965	0.973	0.969
model#5	Train Error	73.548	839.030	0.912	0.969	0.977	0.973
	Test Error	35.177	366.374	0.904	0.966	0.980	0.973

As you can see, these models has very small differences that can be neglected.

This is the ROC Curve of Models:

Comparing to the Second Project

In the second project, the LDA classifier achieved the best results with F-Score of 0.995 on the test data. And as, we can see that designed Neural Networks has relatively lower F-Score than the LDA. However, this does not mean in general LDA works better than every Neural Network in this problem. But there could be a Network that has better outcome than my designs. Hence, we can conclude, on this dataset and with my knowledge and my effort, the LDA has shown a better result.