how can i do a stock market prediction using a neural network?
The effortless way is to use the wet 1 in your head. You use a university to train it instead of a a lot more conventional backprop, although.
Barring that, of course, you’ll find a couple substantial challenges here. Initial, you should recognize and comprehend the characteristics from the ANN approach. You have to understand how large of a network you are talking about.!!!. three layers is all fine and good, but you are going to need to know how a lot of nodes to operate with. The most considerable issue, although, is the inputs.!!!. to predict rates with any kind of accuracy, you’d require to be capable of track, input and train on at least some variables that affect value.
The purely numerical ones (the day of the week, yesterday’s cost, the index costs, etc) are fairly straightforward, but you’ll need to have to account for everything else, too.!!!. people’s minds can change for totally unrelated occurrences, they may blindly follow trends (meaning you will need to have to loop some of your output back) or go against them. They might buy or sell in big quantities mainly because of what analysts say on TV, whether or not it makes sense. You may be capable of model a great deal of that with statistical models, but the devil is in syncing the timing to reality.
Bottom line, it’s an interesting difficulty. Solve it well, and you will never require to operate again 
May 26th, 2010 at 2:50 pm
The easy way is to use the wet one in your head. You use a university to train it instead of a more traditional backprop, though.
Barring that, of course, there are a couple significant challenges here. First, you need to recognize and understand the characteristics of the ANN approach. You need to know how big of a network you’re talking about… three layers is all fine and good, but you’re going to need to know how many nodes to work with. The most significant problem, though, is the inputs… to predict prices with any kind of accuracy, you’d need to be able to track, input and train on at least some variables that affect price. The purely numerical ones (the day of the week, yesterday’s price, the index prices, etc) are fairly easy, but you’ll need to account for everything else, too… people’s minds can change for totally unrelated occurrences, they may blindly follow trends (meaning you’ll need to loop some of your output back) or go against them. They may buy or sell in large quantities because of what analysts say on TV, whether or not it makes sense. You may be able to model a lot of that with statistical models, but the devil is in syncing the timing to reality.
Bottom line, it’s an interesting problem. Solve it well, and you’ll never need to work again
References :