What should a dataset look like for time series forecasting?

Can I do time series forecasting with a dataset that contains apartments from ad sites obtained with:

  • web scraping from 2018 to 2021
  • 13 features
  • Date, Region, area, rooms, level, price, levels, type etc.

This the dataset link for further information about dataset dataset link

Or should I have the historic of the change of price of these apartments over time?

Like an appartement in region 1 with 1 room costs X in 2018 and costs Y in 2019?


2 Answers 2


What you described is a longitudinal panel dataset rather than timeseries data. As such you use methods for panel data regression, mostly fixed- or random effects. Good resources here and here.

  • $\begingroup$ I don't understand your answer neither the ressources that You posted in your answer, should i convert m'y dataset to do future forecasting ? $\endgroup$ Jun 16, 2022 at 21:14
  • 1
    $\begingroup$ As it’s currently written, your answer is unclear. Please edit to add additional details that will help others understand how this addresses the question asked. You can find more information on how to write good answers in the help center. $\endgroup$
    – Community Bot
    Jun 17, 2022 at 3:24

Timeseries datasets should first be representative enough of the business process they are dealing with.

For instance, some processes are better represented day by day, others, hours by hours, etc.

That's why you should first know the objective that will define the neccessary frequency and features, you cannot take all features and time values and get a good result: you should define at least one objective first, analyse your data, define which features seems meaningful, apply correlations functions to know the links between features, etc.

Once you have a good understanding of your dataset, you can start with a few features to see if your model work.

In your case, the best is to apply both solutions separately:

  • Times series to detect how the market is changing and predict prices according to a few time-based features.
  • Multi-dimensional prediction on most recent data in order to predict prices (or other things) using different features.

But you cannot have both solutions at the same time for a beginning. It is better to dissociate them using a few representative features, in order to get good results, and then increase complexity by crossing many feautures.

  • $\begingroup$ But there solutions like multivariate time series, arima and fbprophet with regressors , to predict prices not only according to time but there other characteristics to adjust the price prediction, my solution is to firstly convert my dataset to time series dataset and after that apply multivariate time series forecast, we can at the same time detect how price changes over time according to other features $\endgroup$ Jun 17, 2022 at 9:59
  • $\begingroup$ Could you give an example of raw data you have editing your initial question? Converting data to time series is only possible assigning features to their date, hour, etc. If they are available, it shouldn't be a problem. $\endgroup$ Jun 17, 2022 at 12:05
  • $\begingroup$ dataset has 13 fields. date - date of publication of the announcement; time - the time when the ad was published region - Region building_type - Facade type. 0 - Other. 1 - Panel. 2 - Monolithic. 3 - Brick. 4 - Blocky. 5 - Wooden object_type - Apartment type. 1 - Secondary real estate market; 2 - New building; level - Apartment floor levels - Number of storeys rooms - the number of living rooms area - the total area of ​​the apartment kitchen_area - Kitchen area price - Price. in rubles $\endgroup$ Jun 17, 2022 at 15:47
  • $\begingroup$ This is the link of the dataset on kaggle dataset link $\endgroup$ Jun 17, 2022 at 15:48
  • $\begingroup$ You have the date of publication, consequently you can apply a time series logic on some features like the market evolution on houses with 5 rooms. But you what is your objective? The features you select depends on it. $\endgroup$ Jun 17, 2022 at 15:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.