Regression - Predicting How Much or How Many
At the start of this series on AI for business leaders, I shared that AI can solve five broad categories of problem. In the first three articles, I helped establish a foundation of understanding for us to be able to now explore the first problem category “How much or how many”.
In Machine Learning (ML), the question of "How much or how many?" is called Regression Analysis. Regression Analysis or Regression, as it is more commonly known, is taken from the mathematical field of Statistical Modelling. Statistical Modelling provides much of the theory upon which ML Regression algorithms e.g. Linear Regression are based, however statistical modelling focusses on inference (the relationship between variables) whereas ML focusses more on predictions (outcomes). This distinction between Statistical Modelling and Regression is a gross simplification, but useful for those readers wondering what's the key difference between historical approaches to prediction and those provided by ML. If you are interested to understand the similarities and differences more deeply, this article provides a good summary.
Let’s now look at two simple but useful examples that can be solved using regression techniques. You should then be able to draw parallels within your own organisation and start to explore how ML can be used to solve your "How much or how many" problems.
Regression for Sales Price Prediction
Regression modelling can be highly effective at predicting the sales price in a free market. Estimating the actual sale price provides the seller with an objective measure of value before a house is sold. It also helps the selling agent curate a sales and marketing strategy to maximise the sale value. For those readers who have either bought or sold property, you know that the decision to buy is complex. We consider variables such as size, number of rooms, number of bathrooms, whether it has a garage, the property type and of course location. Location is often one of the most important features as it captures proximity to schools, public transport and shops.
Regression models are not limited to just the properties of the house, they can also take into account environmental parameters (e.g weather, consumer confidence indices or pandemic alert levels) each of which can have a significant impact on the sale price. Being able to predict the impact of these variables in advance allows a seller to make changes such as choosing a suitable day for an auction, renovating or converting a storage room into an additional bathroom.
Regression for Demand Forecasting
Another type of “How much” problem is demand forecasting. All businesses sell products and services with sales that vary over time. For some businesses, such as e-commerce, media streaming and passenger airlines these variations can be significant. Accurate sales forecasting can help organisations set better sales targets, prices and prepare resources to meet demand. When the problem you are trying to solve involves predicting a variable that changes over time (sales in our case), it's called Time Series forecasting. When using ML to solve Time Series problems, we first build a model (brain) based on historical sales data and then use the model to predict into the future.