Beam Category Importance: Find out which types of events impact your business the most

Published on August 28, 2023
Andrei Bizovi
Product Manager

What is category importance?

Feature importance tests are a staple of a data scientist’s wide and varied skill set. They help inform which features or  inputs to a machine learning model move the needle, and by how much. Knowing this, a data scientist is able to optimize the feature set by reducing bloat and noise through pruning less important features. Not only that, but feature importance tests add a vital element of explainability and interpretability.

We put our own spin on feature importance tests with what we call category importance, which is a measure of how much an event category impacts demand at a particular location. At PredictHQ, we provide access to event data, which comes in many categories: sports, concerts, conferences, expos and severe weather just to name a few. 

Through our own research and customer collaborations, we have confirmed that event categories impact different locations and different businesses…well differently. iImagine you have two hotels: one near a college town and another right in the middle of a business district. 

The hotel near the college town is more likely to be impacted by the footfall brought on by college sports and you can expect to see an increase in bookings when those events occur. On the other hand, the hotel in the downtown business district that’s surrounded by offices and showrooms would probably see less if no impact from the college sports games, but they may see an increase in bookings due to conferences, as they tend to be smaller-scale events geared towards business professionals.

We’ve already started uncovering insights that a decision-maker could use to improve operational performance. But what if you could arm a machine learning model with this knowledge? With PredictHQ’s Category Importance analysis you’ll be able to run your own demand data through such tests privately and securely to find out how the different event categories impact each of your business locations. These results can be used by both humans and machines to make better decisions.

How do we determine what events impact you most 

To run an analysis, an event coverage threshold is set at the category level to ensure categories are actually relevant to locations (i.e. ensuring there are enough events near the location for the analysis). This gives all locations a fair chance of achieving importance.

Some industries have more stable proportions than others, meaning they don’t fluctuate as much across population densities and the pattern is usually quite clear. Some explanations for this include having:

  • A good sample size

  • A wide geographical coverage of locations

  • A homogenous industry e.g. accommodation vs. retail (consisting of different retailers)

Averaging proportions across population densities helps to stabilize the trend and uncover underlying patterns. This also means the results from higher density locations contribute more to the proportion, as the population density threshold is a minimum that needs to be met.

For consistency, where possible and available, customer demand is usually:

  • Restaurants: sales, dine in/non-delivery

  • Accommodation: revPar

  • Retail: sales across all products

  • Parking: transaction amount

How to complete your category importance analysis in 4 simple steps

Category importance analyses can be performed either via control center, our interactive web application, or in your own environment via our comprehensive suite of APIs.

Regardless of which method you pick, you’ll need to have a sample of your demand data (sales, units moved, revenue, guest count, etc.) in order for the machine learning model to start correlating event impact against it.

Step 1: Get a sample of your demand, demand indicators or key performance metrics. This needs to be specific to a location – such as a single store, hotel, restaurant, or other places of business. We recommend more than 12 months worth of data for the best results, but the model will work with as little as 6 months of data.

Step 2: Create a category importance analysis In Control Center, where you'll find Beam, our automated correlation engine. 

This analysis automatically checks for all events happening near your location within the given timeframe, and works out the observed impact on your demand at an event category level.

Step 3: You can take these results and request a set of features via our Features API, which provides forecast grade features that can be plugged into your machine learning model. Or you can use these results to adjust your operations ahead of impactful future events.

Reveal your demand catalysts today

With insight into the types of events that matter most to your business, you can adjust operations to align with incoming levels of demand – such as by ensuring enough staff on the schedule and stocking enough inventory to provide a high level of customer service, even during peak demand periods. 

Sign up today to run a category importance analysis on your different business locations to see impactful events within proximity to them.