At its most basic level, an algorithm may be defined as a set of rules for solving a problem in a finite number of steps. But isn’t that just what a computer does? No. If we define a computer as a set of rules written by a human being, then an algorithm may be defined as a computer that writes its own set of rules. There’s a world of difference.
Most of us are aware of the existence of algorithms. Indeed, it’s impossible to escape them: in the digital age, algorithms impact upon just about every area of our lives – from how we make buying decisions through to our ability to obtain credit or even find a compatible partner. While Facebook uses algorithms to find us friends, and Google uses algorithms to scan the web for our image.
The developer first identifies what is known as a ‘seed set’. This is the basic information that is first fed to the algorithm to enable it to function and learn.
The algorithm then processes the information and begins making its own rules. For instance, a risk assessment algorithm used to calculate insurance premium rates for car drivers might say to itself: ‘Ah ha, I see that Dave Smith has a good credit history.’ Research has shown that people with lower credit scores tend to be higher risk drivers. It also registers that Dave Smith is over 25 years old; once again this works in his favour, as older drivers are shown to have fewer accidents. The algorithm also notes that Dave Smith is married, meaning that statistically he is less likely to be a careless driver; he also has a good driving record and has not made any previous insurance claims, so that’s another tick in his favour. He has no driving convictions; once again, the algorithm makes the decision to add this to the mix, as it has determined that this is another safety factor. However, the algorithm now notices that Dave Smith lives in a high crime area, so this is now factored in to the equation, as it moves him higher up the risk register…and so on…with the algorithm continuously learning and refining its processes as it goes.
The developer reviews the results to determine whether the ‘rules’ (e.g. ratings) are working as they should, and making any necessary changes. For example, perhaps the algorithm is failing to account for some vital piece of information, such as the fact that the insured is driving a make and model of car that is considered to be a high risk, so this is then added to the seed data and fed back to the algorithm. The algorithm then creates a rule to account for the new information.
Across the insurance sector, algorithms are now being applied to many more tasks than risk assessment and ratings. Consider for a moment everything you and your colleagues know about your business – what a wealth of information that would be. Now what if all of this wealth of information could be developed into an algorithm and used to create a set of rules? And what if by machine learning this algorithm was able to learn from these rules and build upon them? The possibilities are endless. Around the world, algorithms are quite literally transforming the way organisations operate, enabling them to make faster and better- informed decisions, saving time, creating opportunities and boosting profits.
Want to go deeper? For data professionals and aspiring data professionals, the guys over at Think Big Data have created an infographic showing the 12 most important algorithms that they believe should be in the repertoire of every big data scientist.
Underwriters: save a significant amount of time with our predictive insurance analytics system.
Book a demo to see for yourself how you can save up to 90% of your team's time.
Find out more about the GIROUX platform and how you can attract more capacity, save time and drive profitable growth.