I am a newbie to Genetic Algorithm. I am trying to predict the pattern of occurrences of rules. For example, I have a set of rules defined as below.
Rule 1,
Rule 2,
Rule 3,
Rule 4,
Rule 5,
Rule 6,
For a given date, I could have only Rule 2, Rule 3 and Rule 6 are used. So I would represent this data as a string as stated below
0 1 1 0 0 1
where 1 denotes that the rule is used and 0 denotes that the rule never get used on that day.
So I would have set of data for 5 days as below
011001,
100010,
110011,
101010,
111100,
What I would like to achieve here is to predict the the 6th day data. I have been reading about Genetic Algorithm and Back propagation method to achieve this. I am failed to map my problem with GA or BP due to lack of understanding about those concepts.
I would appreciate if someone could point me to the right direction to help me to map my problem with either GA or BP. Any help is much appreciated.
The occurrences of rules are purely
random
In that case there's no way to predict them I am afraid!
In case the above is not accurate (rule occurrence is not purely random), do you have a training set? How big is it? You should be looking at pattern recognition techniques here more than GAs.
For example recurrent networks seem to be a good fit for your problem. Have a look at this paper, they predict binary time series instead of binary strings but its as close as it gets!
Another approach that comes to mind could be to combine neural networks + GAs in a fashion similar to the way they're doing it on this paper here for financial prediction.
But I am guessing you need a much bigger training set either way, and you'll have to adapt it to your case.
Beware: this is not a trivial task!
GA's are more suited for optimization problems than prediction. If you are interested in using GA's however, you could use it to optimize the parameters for a Neural Network which could be used to predict a pattern. Another useful thing to look at is machine learning using linear regression. With linear regression a regression line can be used as an estimator for predicting the patterns.
You can optimize your rule set by using GA and then provide the optimize set as input to Neural network for prediction, I am afraid you cannot use GA for prediction, prediction required a Inference rules or a well formed training data as input to NN(past information).
