An entire information on intermittent time sequence forecasting in Python with a capstone undertaking
Intermittent time sequence, or sparse time sequence, is a particular case the place non-zero values seem sporadically in time, whereas the remainder of the values are 0.
A typical instance of spare time sequence is rainfall over time. There might be a variety of consecutive days with out rain, and when it rains, the quantity varies.
One other real-life instance of intermittent sequence is within the demand of slow-moving or high-value gadgets, akin to spare elements in aerospace or heavy equipment.
The intermittent nature of a while sequence pose an actual problem in forecasting, as conventional mannequin don’t deal with intermittency effectively. Due to this fact, we should flip to alternate forecasting strategies tailor-made for sparse time sequence.
On this article, we discover alternative ways of forecasting intermittent time sequence. As at all times, we discover every mannequin theoretically first, and implement them in Python.
As at all times, the complete supply code is obtainable on GitHub.
Be taught the newest time sequence evaluation methods with my free time series cheat sheet in Python! Get the implementation of statistical and deep studying methods, all in Python and TensorFlow!
Let’s get began!
Croston’s methodology is likely one of the most typical approaches to forecasting spare time sequence. It typically acts as a baseline mannequin to guage extra complicated strategies.
With Croston’s methodology, two sequence are constructed from the unique sequence:
- A sequence containing the time durations with solely zero values
- A sequence containing time durations with non-zero values
Let’s contemplate a toy instance for example that. Given the spare time sequence under:
Then, based on Croston’s methodology, we create two new sequence: one with non-zero values, and the opposite with the time frame separating non-zero values.
