Placing 51/1121 in Kaggle
Time Series Forecasting Competition

Hi Everyone! Today, I placed 51st amongst 1121 participants in the Playground Series 3, Episode 19 competition on forecasting sales in various countries for different Kaggle products.
This was an artificially generated data competition, but nevertheless, quite challenging due to the strong competition around. I competed against several professionals, student and researchers, most of whom had a lot more experience than me, since I had never done anything related to Time Series forecasting in the past.
I was able to learn by reading some textbooks on Time Series forecasting, and learnt about several novel techniques which I employed in the competition.
However, like most other users, my code was doing poorly on the tests. It seemed that there was a problem. Some experienced users found that there was a data generation error, which caused all the time series to unexplicably converge to the same mean in the test data.
Using this knowledge, I had to find a good transformation to get my predictions back to this mean. While this took quite a lot of submissions, I was able to get a good score, which propelled me to the 51st position (Top 4%)
I had a lot of fun in this competition, and learnt a lot. If you want technical details of my solution, you can read my Silver medal solution post in Kaggle down below:

My Solution Post on Kaggle