Blind Spot in Data Science Training

In recent days, I notice there is this blind spot with regards to beginners, besides the "complexity trap" that I keep talking about. And this blind spot is perpetuated, unfortunately, by the many boot camps or courses out there.

What is the blind spot, you might ask? That is using one machine learning model for EACH problem statement. Is that true? This is why any data scientist or data analyst needs to think about the business problem first and then convert it to a math problem. This step is VERY important but unfortunately not easy to teach in any courses and then that blind spot is further perpetuated by inexperienced instructors.

Let me give an example. My mentees are usually asked to work on coupon marketing data. They start tackling it by clustering, to form customer segments by purchase behavior. But the weird thing is since they know that the behavior is different for each of them, they still went ahead to build A SINGLE cross-sell model for ALL the customers, instead of ONE MODEL for EACH cluster. Having one single cross-sell model for each cluster allows us to capture the intricacies of each cluster since their behavior is different as compared to building a single model for all the consumer segments, thus averaging out any potential signal i.e. a signal in Cluster A may be canceled out by another signal in Cluster C.

In the outside world...things are different. The examples in the courses are structured simply for the learner's sake, to teach the necessary in the limited time. More importantly, we have to understand our tools well and deploy them according to circumstances, rather than sticking to the process taught in courses.

All surgeons know what are tools available to them, but what makes them impactful to people is how they used the tools to save lives, which one to use, when to use, etc. Remember, at the end of the day, data scientist or data analyst earn their keep by solving problems using the tools at hand, not about how many tools we know or how complex the tools we used are.

How can one improve this aspect of problem solving then? I go back to the point about having a hands-on project portfolio and getting a mentor to guide you on it. The more business problems you work on trying to solve with data, the better you can get. Also read up on how other people solve similar problems if possible. It can help to expand your horizon.

I wish you all the best in your learning and job hunting journey! :)

Do check out my other blog posts. Keep in touch on LinkedIn or Twitter, else subscribe to my newsletter to find out what I am thinking, doing or learning. :)