Hi ,
Think of the last decision you made.
It might have been what you had for lunch today, or perhaps what you will do this coming weekend.
Was the decision a one-off?
Are you sure?
Because as long as you continue to live and interact with the world, very few decisions are truly one-off.
Your choice of what you eat for lunch today will influence your future lunchtime decisions because today's lunch decision will provide you with new information you will use next time you need to decide.
Your choice of what you do this weekend will influence your decisions around future weekend
activities for similar reasons.
Every decision you make is part of a sequence of decisions because every decision provides you with new information and time never stops.
So, when you're building a predictive model to assist people with decision-making, what
assumptions are you making about those decisions?
Here's the thing...
Most predictive models fail to take into account the sequential nature of decisions.
In fact, most supervised learning algorithms are based on the assumption of independent decision-making - an assumption that never truly holds in the real world, resulting in the inevitable deterioration of model performance over time.
The only way to truly avoid this issue is by explicitly allowing for the
sequential nature of decision-making. And there is an entire branch of data science specifically devoted to this.
I recently had the opportunity to speak to Prof Warren Powell, a world expert in sequential decision analysis who now applies his research in this area to real-world problems.
You can listen to our conversation at the link below: Episode 36: Sequential Decision Problems
Talk again soon,
Dr Genevieve Hayes.
