10 Time Series Forecasting Methods We Should Know
From product sales, customer demand, inventory forecasting to internal ledger/transactions data, time series datasets are very common within enterprises and easier to be used for forecasting purposes w.r.t. other machine learning problems.
In this blog, we’ll use a simple sales time series dataset to predict the future sales using 10 well known univariate time series prediction methods.
Time Series Types
- Univariate Time Series: only one variable is varying over time.
- Multivariate Time Series: multiple variables are varying over time.
Dataset
We have downloaded a simple univariate sales of shampoo dataset from here. Instead of going deep into the algorithms or mathematical details, we have limited our discussion on using the respective libraries and methods only. We haven’t included further model optimization/hyper-parameter tuning which can be part of further detailed discussions.
Data Characteristics
After plotting the time series graph we should look for the following three characteristics and accordingly we may need to take the right measures or choose the right methods.
- Trend: the values are increasing/decreasing over time.
