Various time-series forecasting algorithms for 1-D data in python. (For code description see below)

In particular, the algorithms implemented/compared are: AR(I), Ridge Regression, Lasso Regression, RandomForestRegressor (those three use sklearn) and LSTM (using Keras). Hyperparameters are only optimized via gridsearch at the moment.

Here I explore some time-series prediction algorithms on four different, real data sets:

• monthly numbers of observed sunspots
• time-series of a chaotic dynamical system (Mackey-Glass)
• atmospheric CO2 concentration.

I find that the methods' performance is highly dependent on the data and the assumptions and problem specific knowledge are thus vital for achieving high performance.

• time_series_1.py: contains all the forecasting functions
• ts_forecast_plots.ipynb: explores the above functions on the different datasets for 1-step and multi-step forecasts.