Despite being a long-standing field, time-series forecasting hasn't been at the forefront of the recent machine learning boom. This course aims to bridge that gap, combining the latest advancements in machine learning with classical statistical techniques to show you how to make powerful, accurate predictions at scale.
Focusing on practical applications and hands-on projects, we’ll teach you the techniques and best practices that supercharge applications like causal analysis, demand intelligence, and workforce planning. If you’re a data scientist or ML practitioner who wants updated knowledge on state-of-the-art forecasting methods, this is your course!
To celebrate our new course, let’s kick things off with a brief introduction to time-series forecasting.
Time-series forecasting is one of the most common and valuable applications of machine learning, but it rarely gets the spotlight in conversations about AI. And that’s unfortunate – because with vast amounts of data being generated every day, time-series forecasting is one of the best ways for organizations to extract meaningful insights and make informed predictions about the future.
In fact, the ubiquitous need for time-series forecasting – not to mention its diverse applications across industries like finance, economics, retail, energy, healthcare, and more – make it a critical skill to add to your career toolkit. For aspiring business analysts, data scientists, or ML practitioners, a well-rounded understanding of time-series forecasting is the jumping-off point for contributing to important decision-making processes and helping organizations solve critical business problems.
In this post, we’ll walk you through an overview of time-series forecasting, the basic process when machine learning is involved, and the interplay between statistical and machine learning techniques. Let’s get started!
What is Time-Series Forecasting?
A time series is a sequence of data points that are recorded over a period of time. Time series data can be collected at regular intervals, such as every day, every week, or every month. Examples of time series data include stock prices, weather data, and sales figures.
Time-series forecasting is the process of using historical data to identify patterns and trends and make predictions about future values and events. It is one of the longest-standing applications of machine learning, widely used in a diverse range of industries today.
What are the Applications of Time-Series Forecasting?
One of the strengths of time-series forecasting is its versatility. It can be used to make predictions about a wide range of events, from product demand to the weather forecast. To see just how versatile time-series forecasting is, check out our blog post, 20 Exciting Use Cases for Time-Series Analytics.
From Classical Statistics to ML Models: A (Very) Brief History
In the past, forecasting relied on classical statistical techniques like exponential smoothing and ARIMA models. In recent years, though, machine learning techniques like deep neural networks and recurrent neural networks have become popular for time-series forecasting. These techniques have the advantage of being able to analyze massive quantities of historical data and learn complex relationships that other forecasting methods might miss. On the other hand, classical statistical techniques have the advantage of being well understood and easy to interpret.
Combining machine learning and statistical techniques gives you the best of both worlds: the interpretability and scalability of time-tested statistical techniques, and a model’s ability to learn complex relationships and work across multiple data sources with relative ease.
How Does Time-Series Forecasting Work?
In machine learning models, time-series forecasting involves the following steps:
1. Data Collection: Your data can be collected from a variety of sources, including databases, PoS systems, spreadsheets, and sensors – it all depends on what you’re forecasting. In some cases, like disease or economic forecasting, you can also introduce publicly available datasets.
2. Data Preparation: No surprises here. Just like in any other machine learning project, the data you collect must be preprocessed and cleaned to remove outliers, missing data, and other anomalies. You’ll also standardize your data at this stage so it’s in a consistent, useful format.
3. Model Selection: Next you’ll select an appropriate model for time-series forecasting. Several models work well for this purpose, including autoregressive integrated moving average (ARIMA), exponential smoothing (ES), and ensemble models. In our course Forecasting with Machine Learning, we’ll help you explore the best options and their applications.
4. Model Training: Like Step 2, this step should look pretty familiar to people who have worked on machine learning projects before. For forecasting use cases, the goal is to find the optimal parameters to use for your model.
5. Model Evaluation: After your model is trained, you’ll evaluate it using a hold-out data sample that wasn’t used in the training of the model. If your model can accurately predict the values in the hold-out sample, you’re in good shape. If not, some troubleshooting and fine-tuning is in order.
6. Forecasting and Ongoing Training: Congrats! Your model can now be used to make predictions about future events and values. Monitor your model’s performance and continuously train it using the most timely data, because cycles and trends can shift drastically in a short period of time. In the case of trends and product demand, for example, it’s likely that your pre-pandemic data looks nothing like the post-pandemic data – so it wouldn’t yield accurate predictions about today’s landscape.
Put Your Knowledge Into Practice with Forecasting for Machine Learning
Intrigued? We’ve barely scratched the surface! In our course Forecasting for Machine Learning, we’ll explore this topic in depth, teaching you how to combine hard-hitting statistical techniques with hands-on machine learning projects. By the time you’re done with this four-week course, you’ll have a working model ready for deployment into downstream applications, ready to tackle some of the most pressing use cases today.
If you're ready to unlock the power of time-series forecasting, sign up today. We’ll see you on March 13!