Join Alan Simon for an in-depth discussion in this video Adding depth and breadth to enrich our analytics, part of Foundations of Business Analytics: Prescriptive Analytics.
- View Offline
- We've already seen how datacore relation is a key aspect to being able to progressively enrich the business impact of our analytical models and also the prescriptive actions that come out of those models. We also saw that in some cases we may need to bring additional data into our models to expand it deep in those business insights that we wanna draw or the questions that we wanna ask, and then have our analytical systems answer. Let's take a look at an example of a state-level department of transportation, and specifically the agency within that department that's responsible for road safety.
As you see here their high level mission is broken down into more detail specific responsibilities such as understanding how weather situations impact road safety, managing speed limits, doing the long-term planning for road improvements based on how traffic loads are projected to increase over the next however many years and also coordinating with the state police and other agencies as needed. Let's say that our road safety agency begins by building sophisticated, predictive, and discover analytical models that primarily use three types of data: road traffic data, accident data, and weather data.
And to get themselves started they begin with three years of historical information from each of these sets of data. Using their models in this initial set of data they can ask critical business questions such as those you see here. For example, they will be able to tell over the past three years at what snowfall levels do the interstate highways within the state become extra dangerous. They can also determine how the safety statistics compare between the interstate highways as well as state roads. They can also look at all the highways within a state and determine if some are safer than others despite having the same snowfall levels.
And if so, what are the factors that make some roads safer than others? They can also ask how much preventative road treatment, such as salting and spraying, before a snowfall actually begins will result in a meaningful reduction in the level of accidents. They can also look at the accident levels during bad weather and determine if anything other than weather-related conditions might also have played a role in high levels of accidents. Now, let's say that as part of the state's planning process during the course of day to day operations, additional questions are raised that actually can't be answered by the models using this initial set of data from these subject areas and with only three years.
What the agency then decides to do is add depth to their data. In other words, for the exact same subject areas that they have, and they're currently analyzing, they simply go deeper. And in this case by adding additional historical data in to the mix. By expanding their data sets from three years to eight years they can run the same data through the same models but still come up with an entirely new set of insights and hypotheses. They can still ask and answer the same questions that we've already looked at, though, with that additional five years of historical data they might actually start to receive different answers.
This in turn spurs an entirely new set of important questions and insights that previously they were unable to ask or answer when they only had three years of data. For example, now they can ask: since the opening of the two new interstate highways within the state over the past three years, what's happened to the overall accident patterns? They can also look at this deeper level of historical information and ask if the primary accident factors during the initial analysis period of three years are the same as they were during the previous five years. And then let's say that the maximum speed limit within the state was changed four years ago.
They can look at the relative accident and safety levels both before and after that speed limit change. So, as we see, we can ask more questions than we were able to ask before and do even deeper analysis but this may not even be enough. Now, let's say that they wanna bring in driver license data and in this case, they bring in 10 years worth of driver's license data. With this additional data they can ask even more questions and this is called adding breadth to the data available for correlation. Now, they can ask if there are any relationships between the number of licensed drivers in the state at any given point and what the relative road safety levels were.
And if so, how that correlated with when the new interstate highways were opened. They can also look at if there's any relationship between drivers who get into accidents during bad weather as well as their overall driving records regardless of weather. Things like tickets and non-weather related accidents. They can also determine if there's any significance between how far away from home someone might be and the likelihood of an accident, and if indeed there is some significance there. If there was a specific time of day or day of the week that happens to have stand outs and then they could take actions based on those insights.
To finish up this example, let's say I add even more breadth to our data. Now, we have eight years of road traffic, accident and weather data. 10 years of driver license data. And we're actually gonna add 10 years of tax return data from our state department of revenue. Now, with even broader and deeper data we can ask an even richer set of questions and gain an even more robust set of insights to drive even more actions. For example, we can look at the relationship between somebody's income, that person's driving record, and then their accident record.
We can ask some very specific targeted questions such as if a single male under the age of 25 with an income less than 25,000 dollars, and who works in one of several specific job categories is more likely to get into accidents at any time or maybe only during bad weather.
- Exploring the analytics taxonomy
- Understanding prescriptive analytics fundamentals and workflow
- Looking at data warehousing and business intelligence
- Exploring big data
- Collecting and processing data
- Exploring triggering events
- Formulating business hypotheses
- Refining and enriching business hypotheses
- Reaching definitive conclusions
- Putting the finishing touches on prescriptive analytics