Try to start as soon as you encounter the competition.

Do exploratory data analysis properly. This will really help you build a good solution.

Try to achieve a base-line model in the first five iterations.

Track your own progress as well as track progress of others by checking out public and private leaderboards.

Try different approaches and be innovative.

Ask appropriate questions in forums but don't waste too much of your time.

If you feel that in order to win you need to form a team, then ask other members to join you.

When you create your team does matter. You need to form a team at least 15 days before the deadline of the competition so your team can work together and come up with the best possible approaches.

The selection of team members also plays an important role. You should team up with those who have a complementary skill set so you can take advantage of each other's knowledge. 

Understand your team member's solutions and discuss approaches in depth. 

Your team members should know each other's strengths and weaknesses so it is easy to collaborate. 

Your team should discuss opportunities--meaning, you need to discuss other possible approaches in order to try them if possible. 

If you lose the competition, don't get disheartened. Try harder next time and learn what went wrong this time. List those things that went wrong and next time you won't make those kinds of mistakes. Try to improve those things. You will definitely win.