Predictive analytics is a complex but necessary technique used in many industries today. To understand and educate people on the topic we felt it was best to consulate our experts on the matter. Dr James McKeone, our Principal Data Scientist, was tasked with explaining predictive analytics in terms of the techniques and models that are used.

What are predictive analytics models

A predictive analytics model is any model that allows you to make a business decision over some future horizon. Typically, a predictive analytics model will have a forecasting or forward-looking component (predictive) and an optimisation or decision-making component (analytics). It’s more than just forecasting sales next 30 days; a true predictive analytics model will then use that 30-day forecast with some rule or algorithm to show the best course of action.

Types of predictive analytics models

There are various models that are used within predictive analytics and include: 

  • Forecast models 
    • This model is one of the most common within predictive analytics. It handles metric value prediction by predicting new data values using historical data. 
  • Classification models 
    • Another very common model used. Similarly, to forecast models, classification models categorise new data using historical data. 
  • Outliers Model 
    • Whilst the above models utilise historical data, the outliers model uses anomalous data within a dataset. It works by focusing on and identifying unusual data. A great model for finding anomalies within data. 
  • Time Series Model 
    • Again, Time Series models focus on anomalies within data and are used to show variable changes over time. 
  • Clustering Model 
    • This model takes data and sorts it into groups based on common attributes and similarities. This is useful for seeing certain data based on a certain group i.e., age, gender, and location. 

With these in mind, Dr James McKeone explains:

“Anything with a forward-looking component (doesn’t even have to be a formal forecast) coupled with a decision element. Could be as simple as whether to carry an umbrella today. E.g., Look at the weather forecast, if it’s >60% chance of rain then carries an umbrella. And it could be as complex as how to allocate your budget across your media channels (BOSCO™).”

Examples of predictive analytics modelling 

Now we have explained what types of models exist, here we give examples for how they’re used across companies and disciplines. Dr James McKeone shares the different models:

“These are some of the predictive analytics models I’ve worked on in the past 5 years or so (some of them just for my own enjoyment, others for a company I worked for)

  • Inventory and stock replenishment models
  • Recommender systems for product substitutes for insurance claims
  • Customer lift time value models -> segment customers expected value then target with marketing
  • Net present value calculations for asset pricing and portfolio optimisation
  • BOSCO™
  • Credit card risk models – forecast who’s going to become a bad debt, reduce the chance of that happening
  • Spatial models for risk-based decisions – e.g., should I continue to loan money to miners at a coal mine nearing the end of its life?
  • Spatial models for sampling data and – Where should we collect more data on the great barrier reef to limit decay.
  • Where should I go on holiday – purchase power parity model for “best” holiday experiences.

BOSCO™ uses predictive analytics modelling to forecast where to spend budgets in both new and existing channels for maximum efficiency. Book a demo to find out how BOSCO™ can help your business.