Many phenomena in our day-to-day lives are measured in intervals over a period of time. Time series analysis methods are extremely useful for analyzing these special data types. In this module, trainee will be introduced to some core time series analysis concepts and techniques.
The module starts with basic usage of Numpy and Pandas library in Python. Then it discusses statsmodels library and its powerful built in Time Series Analysis Tools. Including learning about Error-Trend-Seasonality decomposition and basic Holt-Winters methods. Afterwards it elaborates AutoCorrelation and Partial AutoCorrelation charts and using them in conjunction with powerful ARIMA based models, including Seasonal ARIMA models and SARIMAX to include Exogenous data points.
State of the art Deep Learning techniques with Recurrent Neural Networks that use deep learning to forecast future data points is also discussed. Also, multivariate time series model such as vector Autoregressive. This module also introduces Facebook’s Prophet library, a simple to use, yet powerful Python library developed to forecast into the future with time series data.
Prerequisites : GLM, SFDS, EDA
Having successfully completed this module trainees are expected to be able to:
- Understand and be able to apply the concepts and methods underlying the analysis of univariate and multivariate time series, and the context for interpretation of results.
- Decompose a time series into trend, seasonal and irregular components.
- Understand the theoretical bases of different methods of time series analysis including decomposition.
- Determine how and when to apply different methods of time series analysis.
- Wei, W. S. (1994). Time Series Analysis: Univariate and Multivariate Methods.
- Harvey, A.C. (1993). Time Series Models.
- Chatfield, C. (1996). The Analysis of Time Series: An Introduction.
- Chatfield. The Analysis of Time Series, 7th Edition, CRC Press (2016).
- J. Brockwell and R.A. Davis. Time Series: Theory and Methods, 2nd Edition, Springer Series in Statistics (1991).
- Taylor SJ, Letham B. 2017. Forecasting at scale. PeerJ Preprints 5:e3190v2 https://doi.org/10.7287/peerj.preprints.3190v2
- Da Costa Lewis, N. (2016), Deep Time Series Forecasting with Python: An Intuitive Introduction to Deep Learning for Applied Time Series Modeling, isbn 9781540809087, CreateSpace Independent Publishing Platform
|Topic ID||Topic Title||Lessons|
|TSA1||Introduction to Time Series Analysis||– Exploratory time series data analysis|
– Naive and Extrapolation approach
– Error-Trend-Seasonality decomposition
– Holt-Winters methods
– White noise (WN) model, the random walk (RW) model, and stationary processes.
– Correlation analysis and the autocorrelation function
– A simple moving average
– Exponential Smoothing
– Distributed lag Regression
– Panel data regression
|TSA2||Causality Analysis 1||– Introduction to causality Analysis|
– Granger Causality Analysis
– Causal Inference and Counterfactual Reasoning
– Controlled Regression
|TSA4||Advanced Time Series Analysis||* Optional, recommended after ADM|
– Transfer function models:
– Seasonal ARIMA models and SARIMAX to include Exogenous data points.
– Recurrent Neural Networks that use deep learning to forecast future data points.
– Scalable Facebook’s Prophet time series forecasting
– ANN and SVR for time series forecastings
|– Vector Auto Regressive (VAR)|
– VARIMAX model
|TSA3||Causality Analysis 2||– Difference-in-Differences|
– Fixed-Effects Regression
– Instrumental Variables
(with or without a Randomized Encouragement Trial)
– Propensity modelling***
– Bayesian structural time-series