Dear colleagues,

Working with time-series data, specifically from sensors, there seems to be no consensus, and they are approached differently. Regarding the ideal window size of the time series, it seems to be one of these cases.

Some approaches are covered, such as:

• The ideal window size depends on the knowledge of the study domain, which makes the windowing strategy more empirical;
• Some studies consider the window size with the smallest or largest variance of its values;
• Other approaches seek to remove the trend and seasonality of the time series, making it stationary, and thus test different window sizes;
• There are also complex approaches that use Kolmogorov's entropy and the positive Lyapunov's exponent to show a chaotic time series.

Anyway, I would like to know from colleagues if there are different ways from the ones presented to estimate, at least roughly, an ideal windowing size that can be applied in different time series.

I appreciate the participation of everyone who can contribute, and I hope this edition can help many.

Greetings.