Concept: The AutoFit Algorithm

You might find it challenging to select the most appropriate algorithm for forecasting a given time series. Typically, users go through a trial and error process using each of the algorithms. Then, they compare the forecast results with their corresponding errors and select the 1 with the fewest inaccuracies.

AutoFit

predicts which algorithm to use for a new time series without making you go through the process of trial and error.

Unlike the other algorithms available,

AutoFit

uses a classification model. The classification model leverages a meta-learning framework to identify the algorithm to use based on the features of the time series data. It uses a model that trained on several thousands of time series data.

Once the classifier identifies the features of the time series, it selects and applies a suitable algorithm.

AutoFitt

considers each intersection of data as a time series. So each account on the sheet might have different features. In this case,

AutoFit

applies an algorithm for each account.

AutoFit

selects only from the algorithms available for

Predictive Forecaster

. Example: When you have 80% sparsity in your data,

AutoFit

generally picks the

Croston

algorithm.