Time series forecasting is a predominant field in many of a companies planning efforts. Thus, it is not surprising that slowly, but steadily, Machine Learning is becoming more and more popular in such finance processes.
When tackling a time series problem with Machine Learning algorithms a Data Scientist needs to take certain steps. At the very beginning of each time series forecasting project, the Data Scientists needs to prep and engineer the data. The prep and engineering step comprise fixing date formats, identifying and removing outliers and missing values. Furthermore, time series specific features, such as lagged values, moving functions, trends and seasonality factors can be engineered. This of course produces many - and sometimes useless - features. Thus, the Data Scientist is tasked with identifying the relevant features within the oversized space of all feature transformations. Only then, with the subset of relevant features, the Data Scientist starts training and tuning different algorithms, with various hyperparameter sets.
All those steps are mechanical and repetitive across time series forecasting problems. Thus, we gathered all our knowledge from our previous Machine Learning solutions to time series forecasting problems and created an AutoML tool that automates these steps.