Probabilities, Distribution, Sequence Analysis. Thought Question

Question

I have this idea for a prediction algorithm that predicts a random value with accuracy based a sequence of previous occurrences of an item from the set being chosen and analyse the pattern to improve the accuracy.

So basically an algorithm that accepts two parameters, one being a set of possible choices; the other being an history of those numbers, analyse that pattern and predict the next number in the sequence.

1. Does this make sense to anyone else?
2. If it does, what would be your approach to writing such an algorithm?
3. What language would you refer?

The goal is pattern recognition in overlapping sample sets, find a correlation and establish an equation to predict that with some degree of accuracy.

Answer 1

I have toyed with predicting random sequences, such as what you describe, by applying a Taylor Series Approximation. I'd say that it was randomly successful. Sometimes it worked, and sometimes it did not.

Another approach might be to use a Fourier Transform to find underlying frequencies that might be influencing the sequence, then try making extrapolations on that basis.

Several well-funded attempts have been made to predict the stock market using these ideas, but they were abandoned after they proved to be utterly unreliable.

The old saying about computers is true: Garbage in... Garbage out!