Кто читает:: Департамент больших данных и информационного поиска

Статус:: Курс обязательный

Когда читается:: 2-й курс, 2 модуль

Abstract

The course explores the principles and methodologies of time series analysis and forecasting, with a strong focus on practical applications in business contexts. Such topics as seasonality, trends, stationarity and autocorrelation, using business-oriented examples like demand forecasting, sales analysis and financial modeling will be covered. The course will introduce statistical and econometric models, including ARIMA and MSTL, and advanced machine learning approaches, including Prophet.A special emphasis will be placed on the use of external regressors - such as economic indicators, marketing activity and competitor data - to enhance model accuracy and incorporate real-world influences. Students will learn to integrate these regressors into their forecasting models, understanding their impact on business performance metrics and outcomes.

Learning Objectives

• Understand the foundational concepts of time series data and recognize its unique patterns.
• Identify and apply appropriate models for time series forecasting for business tasks.
• Integrate external regressors into forecasting models to capture complex influences on business metrics.
• Develop skills to evaluate forecasting models and interpret results in a business context.
• Gain proficiency in using software tools (e.g. Python libraries) to analyze time series data and generate forecasts.
• Apply forecasting techniques to real-world business scenarios, improving decision-making and strategic planning.

Expected Learning Outcomes

Being able to plot time series using matplotlib (python library).
Being able to plot time series using plotly.express (python library).
Being able to interpret time series for business purposes with pandas (python library).
Understanding the concept of a time series.
Calculating simple statistics: mean, maximum, minimum, lags.
Understanding stationarity.
Tests for stationarity: ADF test, KPSS test.
Understanding autocorrelation.
Tests for autocorrelation: Ljung-Box test.
Understanding methods of reduction to stationarity and being able to perform them: differencing, log transformation.
Understanding the idea and being able to decompose time series into basic components with various methods: classical, STL and MSTL decomposition.
Understanding AR, MA, ARMA and ARIMA models and being able to forecast with them.
Distinguish ARIMA terms from exploring an ACF and PACF.
Understanding the advantages and disadvantages of using algorithms from the ARIMA family.
Understanding the concept of regressors.
Being able to use dummy variables for regressors.
Understanding and being able to calculate correlation of time series.
Integrating regressors into ML and econometric forecasting models.
Understanding the need for time series preprocessing.
Understanding and being able to use algorithms for detecting anomalies in time series: CUSUM and Chow test.
Being able to treat gaps in data with interpolation techniques.
Understanding and being able to forecast with SARIMA, MSTL, Holt-Winters, Theta and Seasonal Naive with statsforecast (python library).
Understanding metrics to evaluate forecasts quality.
Understanding the advantages and disadvantages of using econometric models.
Understanding regression models for time series forecasting.
Understanding Prophet model and being able to forecast time series with it.
Understanding and being able to tune forecasting models with cross validation for time series and grid search.
Understanding metrics to evaluate forecasts quality.
Understanding the advantages and disadvantages of using ML models.

Course Contents

Time series visualization and interpretation
Time series features
Reduction to stationarity
Time series forecasting with ARIMA models
Use of regressors for time series forecasting
Time series preprocessing
Time series forecasting with econometric models
Time series forecasting with ML models

Assessment Elements

Home Assignments
Exam
There will be a final work at the examination session of the module 2, at the end of December, synchronously with online proctoring at Smart LMS. The duration of the exam is 2 hours.
Quizzes
Short asynchronous quizzes. Each quiz will take 10-15 minutes from starting an attempt and will cover the material of the previous topics. Question types might be a single-choice, multiple-choice or a short answer.

Interim Assessment

2025/2026 2nd module
0.2 * Exam + 0.4 * Home Assignments + 0.4 * Quizzes

Bibliography

Recommended Core Bibliography

Makridakis, S., Wheelwright, S. C., & Hyndman, R. J. (1998). Forecasting: Methods and Applications. Cyprus, Europe: John Wiley & Sons, Inc. Retrieved from http://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsbas&AN=edsbas.F848CE7
Time series analysis : forecasting and control, Box, G. E. P., 2008

Recommended Additional Bibliography

Enders, W. (2015). Applied Econometric Time Series (Vol. Fourth edition). Hoboken, NJ: Wiley. Retrieved from http://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsebk&AN=1639192
Hamilton, J. D. . (DE-588)122825950, (DE-576)271889950. (1994). Time series analysis / James D. Hamilton. Princeton, NJ: Princeton Univ. Press. Retrieved from http://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edswao&AN=edswao.038453134

Authors

Ахмедова Гюнай Интигам кызы

Магистерская программа «Магистр по наукам о данных»

Time Series Analysis and Forecasting

Course Syllabus