Forecasting using neural networks in Matlab

Question

I am doing a project on traffic forecasting for a particular road junction. As a part of data collection process, I have collected the count of number of vehicles on a particular day (From 8 am to 12 am at 15 minute time intervals each) passing through that junction. The following is the time-series after consolidation.

Vehicle Count

2619(8:00 am to 8:15 am)

2925(8:15 am to 8:30 am)

3090(8:30 am to 8:45 am)

3075(8:45 am to 9:00 am)

2786(9:00 am to 9:15 am)

3403(9:15 am to 9:30 am)

3460(9:30 am to 9:45 am)

3122(9:45 am to 10:00 am)

2977(10:00 am to 10:15 am)

3000(10:15 am to 10:30 am)

3081(10:30 am to 10:45 am)

2842(10:45 am to 11:00 am)

2691(11:00 am to 11:15 am)

2816(11:15 am to 11:30 am)

2878(11:30 am to 11:45 am)

2965(11:45 am to 12:00 am)

I am basically trying to develop a simple short term forecasting model using the neural network toolbox inbuilt in MATLAB which should be capable of forecasting the next value in the time series. For instance, if I supply first 15 values of the above mentioned time series as input, the model should forecast the 16th value in time series.

I have the following doubts:

1.) Since I am using a single array i.e. Vehicle Count in my problem, is it ok if I use 'Non-Linear Input Output' option provided in NN ToolBox instead of NAR/NARX?

2.)Suppose I wish to supply first 15 values as the input set, what is the best ratio that can be used to divide the data set into training set,validation set and testing set for the given problem?

3.) How to actually see the forecasted 16th value in NN Toolbox in MATLAB?

I am a beginner in this field. Help is highly appreciated.

Answer 1

If you are predicting a series, with only that series past values as input, then using Nonlinear Input-Output with the series as both inputs and targets is equivalent to using the NAR network.

The NAR network does give you the ability to close the loop, and perform multiple predictions past known data. But otherwise the two networks operate the same.