Hello Guys,

I'm new to the world of AI development, so i'm going to claim complete naivity when it comes to just about everything.

I'm an experienced web developer, so the world of programming and object oriented methodology isnt too distant from my mind. When I say 'web developer' I mean proper web applications, none of this phaphing with building small websites.

I've been working on a webbased app for a while now which works with a large amount of statistcal data, and i'm now looking to develop a AI engine to take this data and analyse it with predicted figures, warning thresholds and things like that.

I've done a little research over the past week into 'tree' 'leaves' and all the usual 'if elseif' type methodology and rule based AI.

I'm really looking for clarifiaction of what AI methods i should be apply to my task and i can then get my head buired into some books on the subject and plan it all properly.

As a rough synopsis of what i'm trying to achieve;

I have bunch of sales data, all numerical and date stamped.

I want to build an AI that'll watch this data and form its own ideas of natural trends. Then act when the figures break from that normal trend.

This isnt as simple as it would seem, and the thresholds for different products will shift during seasons, so if the company starts selling less christmas tree's after january first, the system doesnt need to get all worried, as it would have looked at the past few years information and noted that this was 'usual' for this time of year, you see?

The system also needs to be clever enough to look at a dataset of pretty much any size and then adapt itself, i dont want to be rewriting the application everytime that there is a small change in the data set.

I'd also like there to be a certain element of it learning as it watches, as the end concept for this app is that it'll be a type of 'auto pilot' system for taking action on the data, so watch a person doing it for a few months, and then be able to take control and act as they would.

I dont want to over complicate this all, but i want it to act appropriatly as new figures come in, so if the daily sales come in and they are lower than it would expect, it can fire off an email to the marketing dept to give them a kick in the bum.

I really need direction on what 'type' of AI i need to be looking at to develope a system that would host these features.

Thanks for any help guys, it'll be greatly appreciated. I must appologise for the poor spelling and grammar, i'm a few beers into my late night evening and the fingers are tired after a hard day of coding.

Thanks,

Rob

You seem like you need to do a lot of numerical predictions and analysis. The best AI system for that is probably artificial neural networks. The simplest of them (meaning the ones further from the true biological ones) such as backpropagation networks, are capable of predicting things like stock trends once trained on a significant amount of test data. Research backpropagation neural networks first. They're the best place to start. These networks are also capable of error detection and data mining. Try that method first. I might also note that ANN's is one of the best approaches to AI since real intelligence comes from real neural nets.

The Intellector