Greetings,

I've been on this challenge for a couple of days, getting to weeks, and it has seemed like a rat race in getting a suitable model. This isn't the first data science competition that I'm taking, but this particular one has made me question my practical knowledge of data science. I have gone through the data preprocessing and using the holy grail of machine learning models, xgboost, and my submission score is still very high - with my best score around RMSE: 4648. Which is terrible compared to the scores on the leaderboard.

After using some preprocessing steps like feature correlation, combining numerical features, imputing missing values, aggregation and features scaling, I still get the outrageous high submission score when I train the model with a Random Forest regressor and an Xtreme Gradient Boosting model.

What I'd like to know, that's if I'm not asking for much, is how to obtain the benchmark score of RMSE: 56 that is on the leaderboard. Because at this point, I'm beginning to question every thing that I have learnt on data science. What am I doing wrong? Is it from the preprocessing or from the model? Cos I know, assuming my knowledge are even valid anymore, these are the two major sources of the poor scores.