That is a rather broad question, and there is tons of literature about quantitative analysis and stock market prediction using machine learning.
The most classical example of predicting the stock market is employing neural networks; you can use whatever feature you think might be relevant for your prediction, for example the unemployment rate, the oil price, the gold price, the interest rates, and the timeseries itself, i. e. the volatility, the change in the last 2,3,7,..., days etc. - a more classical approach is the input-output-analysis in econometrics, or the autoregression analysis, but all of it can be modeled using neural networks or any other function approximator / regression in a very natural way.
But, as said, there are tons of other possibilities to model the market, to name a few: Ant Colony Optimization (ACO), Classical regression analysis, genetic algorithms, decision trees, reinforcement learning etc. you name it, almost EVERYTHING has probably been applied to the stock market prediction problem.
There are different fond manager types on the markets. There are still the Quants which are doing a quantitative analysis using classical financial maths and maths borrowed from the physics to describe the market movements. There are still the most conservative ones which do a long-term, fundamental analysis of the corporation, that is, looking in how the corporation earns money and where it spends money. Or the tactical analysts who just look for immediate signals to buy / sell a stock in the short term. And those quantitative guys who employ machine learning amongst other methods.