I'm a newbie too but this is what I do.Experienced people, please point out the flaws in my answer.
1)Start with GBM/Xgboost.Quick to iterate and usually gives decent results. Then probably compare with other algo performance like SVM etc.If you have time, try using multiple models and their forecast as input in one final model(doesn't always work!)
2)Your data will tell you how to go about it.Check out the target variables distribution along with various input features.Most importantly try to 'see' how they behave.Plot every possible relationship you can think of.
Note:I have only been doing this for three months and would like to know some experienced person's answer to this.