I am currently working on an analysis project where I wish to predict sickness levels of employees in different offices for a company for the remainder of 2017.
In the past, I have made predictions based purely on predictors and also just purely on the time series. For this project, I feel I need to combine the 2 types of forecasting for the must accurate results.
I have the following data available:
- The past 4 years sickness data measured in hours per day per office.
- The number of people working in the company in each office for each day of the year for the last 4 years.
I feel that combining these 2 measurements will lead to reasonably accurate forecasts. I can see that that the time series data has a seasonal element to it and a slowly declining trend. My problem is combining the employee numbers per office with historical sickness data per office.
Any help on this is much appreciated. I am attempting to do this in R but any theoretical advice is appreciated too or a book/website reccomendations for learning this kind of methodology. Thanks!
This is called Dynamic regression and is described very well, with examples, in Hyndman & Athanasopoulos (2016) chapter 9.1.
- Solved – Problem with classification machine learning: prediction is always wrong
- Solved – What are Pros and Cons on using deep learning to predict stock prices with features like time, revenue, profits, interest rates and etc
- Solved – Some questions about quarterly and monthly timeseries
- Solved – Cross-validation techniques for time series data
- Solved – k-fold CV of forecasting financial time series — is performance on last fold more relevant