A week ago there was a discussion about wheter everyone used time-series/forecasting approach or they didn't depend on time related features. I used this baseline notebook in one of my early submissions - I think it can be a good introduction to chech the time dependence. or as a started notebook... note that in order to use time related features you need to inspect the target depence on time (e.g if there is a noticable trend/change over time)
Notebook link: https://www.kaggle.com/code/ahmedalghali/geoai-ground-level-no2-time-series
I'm curious how many of you you include time related features on their approach, and who doesn't.
Great walk through. I included some time components in my analysis. It performed better locally but was very poor on the test dataset. Do you consider time series in your submissions?
Thanks, since it performs poor on the testset that means it's irrelevant/overfittted try to evaluate each time derived feature alone with your basemodel (without time realted features) you will find out there are some features that are harming the valudation accuracy. try removing it and rely only.
I used few time related feature and indeed it boosted the performance.hint: try avoiding plain year, week and day feature they seem to overfit.
Thanks for sharing
C'est une excellente questions. Moi je n'ai pas utlisé les éléments temporels mais les resultats sont à peu pres bon seulement qu'il y a un risque de overfitting. les donnnées sur apprend tellement
I think you're sharing a misleading notebook. I submitted and it gets a poor 14.7xxx score. Do not share what you're not using to put the other ones on a false trail. I hope that it is not deliberate.
I'm sorry if you get it wrong mostml. I'm not using this notebook at the moment I have tried different approaches after if of course, and it is "starter notebook" it is not meant to have the best accuracy or postion you the 1st in the leaderboard, there is not misleading information in the notebook as far as I know, if you see any code errors, something that is not right feel free to tell.
My point was that the shared information seems to be partial concerning the related theme. I'm not waiting for any notebook to get 1st in the Leaderbord 😂.
It's a starter notebook. It is meant to help you with your first submission which you can improve on. You have to try different approaches to improve your score. The fun part is to experiment and figure out what is working and what is not