How to prepare the Input for feedforward neural network?

Hi All,

I have to use feedforward neural network for 2-class classification problem. I have EEG data (of 1s, recorded at the sampling frequency of 1200) for two mental states in the following format: Number of EEG channels x Number of samples/points x Number of iterations = 22x1200x200; As the awake and asleep mental state was recorded for 100 times each.

I'm confused that how i should prepare the input signal for feedforward neural network.?

In most of the literature the input layer is equal to the number of channels, but in the examples of NN database they follow the following format:

Input signal= samples x iterations

Targets= number of classes x iterations

Based on what i inferred from the examples i prepared my data in the following format:

Input Signal= 26400x 200

Targets= 2x 200

Although i'm getting acceptable results, but i am not sure if this the right way to it.?

Answer 1

[Dheeraj Singh]

You can refer to the following example where Machine Learning techniques have been used on EEG data:

https://in.mathworks.com/company/newsletters/articles/using-machine-learning-to-predict-epileptic-seizures-from-eeg-data.html

[Joana]

Yes i considered this example, but he's dealing with the features extracted out of raw EEG data, i'm trying to work with the RAW EEG data, where NN can learn some useful features on it's own.

[Dheeraj Singh]

Its difficult to comment on the data without looking at it.

There are some basic things you can look at such as:

1. For each output there is a signal
2. Each signal can be described as a set of features

If you don't know the features you can use feature extraction to extract the features from the signals. Please refer to the following link:

https://in.mathworks.com/help/signal/measurements-and-feature-extraction.html

The following example gives the entire workflow for ECG signals including feature extraction:

https://in.mathworks.com/help/signal/examples/classify-ecg-signals-using-long-short-term-memory-networks.html