I started doing Kaggle competition seriously since 2021 February and became Kaggle Competition Master on 2022 March. During that 1 year, I have won 2 gold, 1 silver and 2 bronze medals.

Here is my 7 suggestions about how to win a Competition:

  1. Hard work: the competition I participated in usually takes me around 200 hours to get a decent result (Silver +). Definitely, the time needed can be lower with more experience.

  2. Teamwork: teaming up with experienced people can boost the learning process and also get a higher chance to win

  3. Jump out of local optimum: don’t spend too much time on hyperparameter tuning or small model structure tuning but rather put more data on data investigation, feature engineering, and very different model structure.

  4. Good pipeline: preprocess -> model training -> post-process, machine learning pipeline can be complex but having a good pipeline is essential because it allows more experiments and thus gives a higher chance to hit the lucky spot.

  5. Good Cross-Validation setting: Having a good CV setting can be great to find the right direction to go fast. Relying on Public LB can be sometimes quite dangerous and slow.

  6. Curiosity: Having enough interest in the competition is probably the main driven motivation for me to push higher and higher scores.

  7. Learning one topic at a time: don’t get overwhelmed by too much information at one time, just take one discussion post or one kernel to learn. We will catch up in the end with enough time. Playing Kaggle is not like horse racing but rather a research process.

Hope this can be helpful to some extent. Let me know if there is anything else you are interested!