The author shares some important aspects of Applied Machine Learning that can be overlooked in formal Data Science education.
Yes I’ve leaned into a clickbaity title but hear me out! I have managed multiple junior data scientists over the years and in the last few years I have been teaching an applied Data Science course to Masters and PhD students. Most of them have great technical skills but when it comes to applying Machine Learning to real-world business problems, I realized there were some gaps.
Below are the 5 elements that I wish data scientists were more aware of in a business context:
- Think twice about the target
- Deal with imbalance
- Testing must be real-life
- Use meaningful performance metrics
- The importance of scores — or not
I’m hoping that reading this will be helpful to junior and mid-level data scientists to grow their career!
In this piece, I will focus on a scenario where data scientists are tasked with deploying machine learning models to predict customer behavior. It’s worth noting that the insights can be applicable to scenarios involving product or sensor behaviors as well.
Let’s start with the most critical of all: the “What” that you are trying to predict. All subsequent steps — data cleaning, preprocessing, algorithm, feature engineering, hyperparameters optimization — become futile unless you are focusing on the right target.
In order to be actionable, the target must represent a behavior, not a data point.
Ideally, your model aligns with a business use case, where actions or decisions will be based on its output. By making sure the target you are using is a good representation of a customer behavior, it is easy for the business to understand and utilize these model’s outputs.