0

Sorry I'm new to machine learning and statistics.

For time series predictions, do you use RNN or something? For example, the past 2 years' sales of a product.

TBH Im pretty much unfamiliar with how statisticians deal with time series data.

I'm dealing with some time series problems and I'm not sure if I should put all my trust on RNN or some other DL models.

Sorry I'm unable to provide the specific data here, don't know if it will cause vagueness. But the sales data is definitely a topic of interest.

user900476
  • 423
  • 1
    There are entire textbooks dedicated to time series analysis that will not even address modern methods that involve neural networks. Do you have a more focused question on time series analysis? – Dave Feb 13 '23 at 18:26
  • @Dave Thank you! I can't provide too specific data as it relates to my thesis, but you can view it as the sales data as I mentioned. BTW what do you think of the RNN models? do you trust it? I mean if it works fine then I dont have to research on other models – user900476 Feb 13 '23 at 18:30
  • RNNs are actually not very often seen in academic forecasting competitions. The dominant contender for the M5 competition was LightGBM in various flavors and combination with predictors. What kind of sales are you looking at? Retail sales behave very differently from electricity demands per household, or total car sales per year. – Stephan Kolassa Feb 13 '23 at 18:33
  • @StephanKolassa Stock price data tbh but there seems to be a lot of non deep learning methods for it. I just dont have enough time to read them. BTW where is the M5 competition held? is it on Kaggle? Would you like to provide me the link? – user900476 Feb 13 '23 at 18:36
  • Stock prices are yet something completely different, I wouldn't call these "sales". In that case, the M6 competition is probably more appropriate: https://m6competition.com/. It's not on Kaggle, probably because Kaggle does not support rolling origins. But there should be forums on the M6 website somewhere. – Stephan Kolassa Feb 13 '23 at 18:40
  • @StephanKolassa Thank you! it seems I cannot download the models from the leaderboard? maybe because the competition is still ongoing? Would you please tell me where you got the models of the M5 competition? – user900476 Feb 13 '23 at 18:49
  • Probably, yes. The M5 competition was hosted at Kaggle: https://www.kaggle.com/c/m5-forecasting-accuracy. There was an entire special issue of the IJF devoted to it. But it was on retail sales from Walmart, so the applicability to stock prices is dubious. – Stephan Kolassa Feb 13 '23 at 19:32
  • @StephanKolassa Thank you! Im very interested in the retail sales data as well. TBH it (sales data) has something to do with one of my past interviews for a data scientist internship. – user900476 Feb 14 '23 at 04:16
  • If you are also interested in retail forecasting, there is an entire review paper on that in the M5 IJF special issue, by Fildes, Ma & Kolassa, plus a postscript. And if you search for my name and "retail forecasting" on YouTube, you will find a few videos you might find interesting. – Stephan Kolassa Feb 14 '23 at 08:53
  • @StephanKolassa Thank you! I have found your videos on youtube. Very educational I'm very glad I found your opinions on deep learning models in the video. – user900476 Feb 14 '23 at 10:55

0 Answers0