用于时间序列数据预测的线性回归模型选择

Question

I have a signal and want to predict y which present Number of requests, using regression models. Currently, I am using OLS regression model to predict y. But the prediction error is very high, as my signal has a lot of variations (ups and downs) as shown below.

I noticed that my model most of the time overestimate y (Number of Requests), especially if the points to be predicted is preceded by large value of y's. As indicated below in the yellow and red circle.

So I am not sure if there's a robust regression models to accommodate this problem of having a lot of variations in my datasets. Also is there any way to segment out these large values by adapting the window size such that it doesn't include these values?

Could you please advise

Answer 1

From the visualization of the error I would say a linear model is not appropriate and you should consider using something that handles periodic data as well as moving average - your data appears to have periodic elements, and a moving average element that goes beyond something "linear". Consider something like ARIMA. Here's a link to a tutorial on ARIMA: https://machinelearningmastery.com/arima-for-time-series-forecasting-with-python/ Please post the results :)

Vishaal