Demand Forecasting Using Time Series

Overview

Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!

Grab it

This course is the second in a specialization for Machine Learning for Supply Chain Fundamentals. In this course, we explore all aspects of time series, especially for demand prediction. We'll start by gaining a foothold in the basic concepts surrounding time series, including stationarity, trend (drift), cyclicality, and seasonality. Then, we'll spend some time analyzing correlation methods in relation to time series (autocorrelation). In the 2nd half of the course, we'll focus on methods for demand prediction using time series, such as autoregressive models. Finally, we'll conclude with a project, predicting demand using ARIMA models in Python.

Syllabus

A First Glance at Time Series

In this module, we'll get our feet wet with time series in Python. We'll start by getting familiar with where time series fits in to the machine learning landscape. Then, we'll learn about the main types of time series and their distinguishing factors, including period, frequency, and stationarity. After pausing to learn how to plot timeseries in Python, we'll explore the differences between seasonality and cyclicality.

Independence and Autocorrelation

In this module, we'll dive into the ideas behind autocorrelation and independence. We'll start by digging into the math of correlation and how it can be used to characterize the relationship between two variables. Next, we'll define its relationship to independence and explain where these ideas can be used. Finally, we'll combine correlation with time series attributes, such as trend, seasonality, and stationarity to derive autocorrelation. We'll go through both some of the theory behind autocorrelation, and how to code it in Python.

Regression and ARIMA Models

In this module, we'll start by reviewing some of the basic concepts behind linear regression. Then, we'll extend this knowledge to feed into lagged regression, an effective way to use regression techniques on time series. Once we have a solid foothold in basic and lagged regression, we'll explore modern methods such as ARIMA (autoregressive integrated moving average). All of this is building the framework for more advanced machine learning models such as LSTMs (long short-term memory network).