How to approach a machine learning programming competition -

many machine learning competitions held in kaggle training set , set of features , test set given output label decided based utilizing training set.

it pretty clear here supervised learning algorithms decision tree, svm etc. applicable. question is, how should start approach such problems, mean whether start decision tree or svm or other algorithm or there other approach i.e. how decide?

so, had never heard of kaggle until reading post--thank much, looks awesome. upon exploring site, found portion guide well. on competitions page (click competitions), see digit recognizer , facial keypoints detection, both of competitions, there educational purposes, tutorials provided (tutorial isn't available facial keypoints detection yet, competition in infancy. in addition general forums, competitions have forums also, imagine helpful.

if you're interesting in mathematical foundations of machine learning, , relatively new it, may suggest bayesian reasoning , machine learning. it's no cakewalk, it's friendlier counterparts, without loss of rigor.

edit: found tutorials page on kaggle, seems summary of of tutorials. additionally, scikit-learn, python library, offers ton of descriptions/explanations of machine learning algorithms.