“Business needs prediction,” says Siegel. “For prediction you need machine learning. predictions improve all the major work we do, they make all the big operations of our cross-sectoral organization more efficient, so we use machine learning to predictive analytics. sometimes called.”
Siegel explained that while machine learning is widely regarded as a powerful tool for business, such projects often fail during the implementation stage. He attributed this pattern to the lack of standardized business practices for predictive analytics projects, suggesting: Business professionals with at least some technical understanding of machine learning will be more successful. “In other words, experts must be able to answer, ‘What does it mean to act systematically on probabilities over and over again to improve operations?'” he said. In his talk, he provided his six-step hands-on method for executing a machine learning project.
But before step-by-step practices can work, organizations need more people who understand and embrace probabilistic thinking. Are doing — Not “just a data scientist doing analytics in the corner”.
Graph via fivethirtyeight.com. Silver probabilities showed uncertainty, but were widely misunderstood.
But that is easier said than done. He cited the backlash that forecast analyst Nate Silver faced after the 2016 US presidential election as an example of a common misconception about probability. Many criticized his model when Trump won, when Silver gave Donald Trump a 28.6% chance of winning and Hillary Clinton a 71.4% chance. However, Siegel wrote in his presentation: And that is probability. So in situations like this one, it happens 30 times out of 100, or 3 times out of 10. It’s not likely to last long. Silver’s model “expressed uncertainty first and foremost,” he writes. Unfortunately for him, however, his model probabilities were widely misunderstood as definitive predictions. This confusion highlights the importance of clarifying What the model predicts, how successful it is, and how to act on those predictions.
Once business professionals and leaders can metabolize this semi-technical understanding, they can more effectively use Siegel’s six-step approach to implement machine learning projects and understand the process as a change management process as well. Become. Siegel outlined his six steps: establishing deployment goals, establishing prediction goals, determining metrics, data preparation, model training, and finally model deployment. While most of the general discussion centers around his steps in training, he emphasized the importance of truly understanding what each step means, especially the predictions in this context.
Examples of misleading headlines about predictive capabilities of machine learning and AI.
“Unfortunately, the word precision comes up all the time,” he says. “Most human behavior is not predictable with a high degree of confidence. We don’t have a magic crystal ball, and we can’t expect computers to.” It is not necessary for individual predictions to have a very high level of accuracy, but rather to achieve a percentage of correct predictions. Better than guessing. Once obtained, companies can act on that information, resulting in cumulatively better performance and, in turn, higher revenue.
Ultimately, Siegel urged the audience to rethink their approach to machine learning projects. Instead of looking at these efforts as purely technical, we need to look at them as business projects with a technical component. This restructuring enables organizations to better align their technical capabilities with their business objectives for more successful implementations and maximum value.
