It has often been said that a great man exists behind every great data model. Enter Andrei Marius Popescu; a man on a mission to make predictive modelling easy for everyone to understand. We wanted to address a recent topic that has caused a significant amount of controversy. Do incorrect models always indicate a failure, or might something else be occurring behind the scenes? Let’s take a quick look at what Andrei Marius Popescu had to say.
First of all, thank you for joining us. We know that you’ve been quite busy with a number of ongoing projects.
Andrei Marius Popescu. “Although I’m currently in the middle of researching cognitive schema, there’s nothing wrong with taking a break.”
Can you begin by describing the purpose of a data model?
Andrei Marius Popescu. “Data models are used to predict the future outcome (or outcomes) of a specific event with a fair degree of accuracy.”
Are there different types of models that can be used?
Andrei Marius Popescu. “There are numerous possibilities. Typical examples include Bayesian models, Fibonacci patterns, and network-oriented models.”
How do you decide which one(s) to use?
Andrei Marius Popescu. “This depends on what you’re trying to predict. For instance, Fibonacci retracements are often employed within the financial sector.”
Are some models more accurate than others?
Andrei Marius Popescu. “This is the million-dollar question. While the model is obviously important, the data is even more critical.”
Can you give us an example of a time when a model was sound, but the results were unexpected?
Andrei Marius Popescu. “One notorious event involved when the Coca-Cola company decided to change its formula for Original Coke in the 1980s. While the models predicted a favourable outcome, the firm experienced a severe backlash from the public.”
What factors do static models have difficulty interpreting?
Andrei Marius Popescu. “I’d say that human psychology is the most relevant wild card here. Even the most advanced systems can have difficulty interpreting how emotions factor into the equation.”
You mentioned the word “static” in the last answer. Can you elaborate?
Andrei Marius Popescu. “Static models cannot be modified according to specific circumstances. While these can be reliable, they are also limited in terms of how they can be applied.”
Are there any dynamic data models?
Andrei Marius Popescu. “These actually refer back to my research on cognitive schema. Some of the latest digital formulae (such as NoSQL) are coming close to embracing a more dynamic approach.”
Speaking of accuracy, how often are your data models proven correct?
Andrei Marius Popescu. “My accuracy currently hovers just over 75%. I think that this is a relatively good figure when compared to some existing approaches.”
Can a sound model be corrupted by incorrect data?
Andrei Marius Popescu. “Almost always. This is not dissimilar to fuelling a diesel vehicle with petrol. The outcome is hardly favourable.”
Might you be able to provide another example of how such an error can occur?
Andrei Marius Popescu. “Imagine a plane that happens to be fed the wrong altitude data during a final approach. While this is a stark example, I feel that it cements the point.”
Andrei Marius Popescu
OK. So, how does randomness fit into the picture?
Andrei Marius Popescu. “Randomness is part of reality. Radioactive decay, turbulence, and genetic mutations are all relevant here. In terms of data modelling, these essentially represent the ‘X’ factor.”
How can models account for this type of randomness?
Andrei Marius Popescu. “There are several ways that I’ve discussed in previous online publications. Stochastic modelling, and probability distributions have both proven to be somewhat effective.”
You replied “somewhat”. Is there any way to eliminate randomness?
Andrei Marius Popescu. “A recent visitor to my professional profile asked the same question,” Andrei Marius Popescu notes. “Unfortunately, mathematics hasn’t yet advanced to this point.”
Do you think that it will ever be possible?
Andrei Marius Popescu. “Honestly, no. Randomness will always be present. The goal is to mitigate its impact on a potential predictive outcome.”
Do you believe that this type of scenario is a failure?
Andrei Marius Popescu. “Absolutely not. This is all part of the learning curve, and mistakes allow us to hone our skills.”
How might randomness be applied to a sports wager?
Andrei Marius Popescu. “Unexpected weather conditions, a sudden player injury, or a team forced to forfeit due to political events are all examples that have occurred in the past.”
How might a model be modified after an error is detected?
Andrei Marius Popescu. “It could be possible to change the algorithm, to employ AI to obtain more data, or to simultaneously incorporate more than one model during an analysis.”
So, randomness might actually be a good thing?
Andrei Marius Popescu. “In some cases, yes. If it points us in new directions, there’s really no such thing as a failed model.”
