Time Series Prediction with Different Sources

user_13539970 0
Viewed 60

First of all, I'm not experienced in model training so please be gentle :)

I need to make some time series predictions to handle an issue related to my product. If the value is too low my customer ask for a quick fix and they cannot know that until it happens. I have collected some data related to different customers but the same product.

Data_1

time value
2020-09-20 600
2020-09-21 450
2020-09-22 350
2020-09-23 300
2020-09-24 150
2020-09-25 50

Data_2

time value
2020-09-20 50
2020-09-21 600
2020-09-22 550
2020-09-23 400
2020-09-24 200
2020-09-25 50

When the value hits 50, we change the product and it's value goes 600. I tried the prophet and kats from facebook and they predict on training data. What I want is train with data_1 & data_2 & data_3 ... and predict with data_4 that can start from 50-600 depending on customer. What would be your approach?

TLDR

Same product,different sources. Same dates,different values. Cut them from 50-600 points or combine them all? How to approach ML model?

facebook-prophet time-series machine-learning artificial-intelligence python
1 Answers

"Different Customers - Same Product" data gives an insight to you . But each product data (in each customer) may have its own characteristics.

You have to check if data distribution same in each customer. If not, please get the data of the product for each customers and do the math.

For the Time-Series approach, i would say you can try many algorithms in kats and facebook prophet itself and check the results-Backtesting (MAPE is good for your problem). Look at the best one (minimum error) and implement it.

user_129069200
Related