1 · Time series foundations
What makes time-ordered data special, its trend/seasonality/noise components, and handling datetimes and resampling in pandas.
●Track 12 · live
Time Series Forecasting
Predict the future from the past — ARIMA, smoothing, and ML forecasting.
Data with a time axis breaks the i.i.d. assumption every other ML track makes — order matters, and you must never shuffle. This track covers the full forecasting toolkit: trend/seasonality decomposition, stationarity and the ADF test, ACF/PACF, AR/MA/ARIMA/SARIMA, exponential smoothing, time-series feature engineering, walk-forward validation, and ML/DL forecasting. pandas, statsmodels, and matplotlib all run in your browser, so the classical models are real — deep models (LSTM) are read-along.
19 / 19 lessons7 chapters
2 · Stationarity & autocorrelation
The property classical models require, how to test for it (ADF), read ACF/PACF plots, and fix non-stationarity by differencing.
3 · ARIMA family
The workhorses of classical forecasting — AR and MA, combined into ARIMA, and extended with seasonality into SARIMA.
4 · Smoothing methods
Moving averages and exponential smoothing (Holt-Winters) — simple, robust forecasters that are often hard to beat.
5 · Feature engineering for ML
Turn a series into a supervised dataset — lag features, rolling statistics, datetime features — and validate WITHOUT leaking the future.
6 · ML & deep forecasting
Forecast with gradient-boosted trees on engineered features, then meet the sequence models (RNN/LSTM) built for time.
7 · Evaluation & capstone
Forecast metrics (MAE/RMSE/MAPE) and baselines, backtesting with walk-forward validation, and a full end-to-end forecasting project.