At its heart, predictive analytics is about making better choices. If your company has the opportunity to make an informed decision, then it is likely to be more successful in the long run. There are plenty of ways that predictive analytics does this – from artificial intelligence to clustering to regression. One particularly important tool for predictive analytics is time series analysis. We might refer to this as creating forecasts for the future. This is a phrase that we’ve all heard before. In fact, when you read the word forecast, you probably thought of a graph that looked suspiciously like this:
The immediate question is how can we use this information to make business decisions? As it happens there are a few relatively straightforward answers to that question:
- You could forecast demand for a certain product over the next two years.
- You may want to forecast revenue for your company.
- You could forecast seasonal staffing needs.
In fact, while these are the relatively straightforward kinds of questions you could answer, there are plenty more that may be even more useful, though less obvious. This may include everything from predicting when a part in a machine may need to be changed to predicting how employee engagement will change throughout the course of a year.
To truly evaluate the trends, it is important to understand the characteristics that may be present as part of a time series analysis. We’ll look at each of the characteristics of a time series (courtesy of this fantastic overview from Penn State) and consider why these matter for a business.
Is there a trend?
The first characteristic of a time series is also, perhaps, the most straightforward one. Any reasonable business would want to know whether their profit (or revenues, or production, or…basically anything else) is moving up or down. If there is a trend, you’ll be able to see whether the metric that matters for your business is increasing, decreasing, or not doing much at all. Sometimes this is quite obvious and other times…not so much.
Is there seasonality?
The next most important question to understand about your data is whether or not there is seasonality present. Do you see a regularly repeating pattern within your data? This pattern could occur over any time period – a day, week, month, year, or anything else in between.
Are there outliers?
Outliers in a time series analysis are the same as you would find in any other type of model. You’re looking for a data point that is far away from the rest of your data. Perhaps you had a particularly good (or bad) day of sales. It might be necessary to suppress this information to get a truly useful model.
Is there a long-run cycle or period unrelated to seasonality factors?
A long-run cycle is exactly what it sounds like – a pattern in the data that occurs over a long time. Think decades or centuries as opposed to years or weeks. For many businesses this won’t be much of a consideration, so we won’t consider it here.
Is there constant variance over time?
Variance is how much data changes. Constant variance points to consistency in the data. If there is not constant variance in the data, then it could point to volatility. The peaks and troughs in data may be increasing or decreasing. Depending on what you’re forecasting, this could either be good or bad. For example, if you have data that has variance that is decreasing over time, that may mean that what you’re forecasting is stabilizing. If variance is increasing over time, the opposite may be true. Like everything else, context matters.
Are there any abrupt changes to either the level of the series or the variance?
Perhaps your company introduced a successful new product line or you just added in a new algorithm to reduce spam. Your business might see an abrupt change in a time series after making a change like this. Visually, the time series will be consistently at one level before a sudden increase or decrease at which point it will consistently be at another level. Like all other things, this change may represent something positive or negative for the business. Often, this change will either require the company to investigate what happened or will be validation for a change that was already made.
Understanding forecasts will help a business to make better decisions about its future. These six questions can help you to better understand your data and react accordingly. In future weeks, we’ll explore different aspects of forecasts, including how to know if there isn’t even a pattern in the first place.
This post originally appeared on the CompassRed Data Labs blog.