Neural networks model for classification

Hi

i am new to neural networks and just used it for classification on a dataset. As i read that neural networks are particularly well suited for complex decision boundary problems over many variables, i used it for my problem.

my problem can be defined as 40 variables which affect the output class

the dataset has 40 input variables: for eg: 1: air temperature 2: humidity value 3: vegetation density in % 4. radiation values and similar variables which affect the final output class "1 or 0"

1 : indicating heavy rainfall 0 : less rainfall

airtemperature  humidityvalue  vegetationdensity%  radiationvalue  Class
32            54             80              32             1
34            76             90              54             1
12            23             21              11             0
22            74             10              82             0

So i have run neural networks on a dataset and it gave me around 91% accuracy over the trainig set and 86% accuracy over the testing set.

Can I directly use this neural network model to test on another dataset?

To explain more my question is : If i have similar another data set for a different region, will I be able to predict how the results will vary?

Thanks!

[Greg Heath]

"Similar other data" is to imprecise for any one to give you more than the following answer:

It depends on how similar the data is.

Greg

[santz venu]

What I meant by "Similar other data" is testing on another dataset which has the same features.

Let me explain :

Say I have a dataset which has all these variables /parameters for city Hongkong.

I create my neural network model by using this dataset of readings obtained for honkong city . So now I have my model ready.

Now if I want test my model on a dataset having the same variables for differnt city say Tokyo.

So since my city is different my variables/parameter values will surely vary.

This dataset of tokyo will still have the same variables but since its for a differnt city the readings will vary. This is what I mean by similar other dataset

My question is What risks will my model have if I use my "honkong based" neural network model for testing on the tokyo dataset.

Hope my question is clear?

[Greg Heath]

Your question is clear. However, it cannot be answered without a quantitative comparison of the two sets of data.

It might be helpful to try to reduce the dimensionality of the input. I would try PLSREGRESS before STEPWISE, STEPWISEFIT or SEQUENTIALFS.

Hope this helps,

Greg

[Greg Heath]

SORRY. I meant to say

I would try STEPWISE, STEPWISEFIT or SEQUENTIALFS before PLSREGRESS

Hope this helps.

Greg