The Houseplant Dilemma - Navigating The Most Common Pitfall in Data Science

September 20, 2023  ·  2m read

We’ve all been there - despite your best efforts and countless promises, you forgot to water another houseplant and it’s starting to look brown and kind of crispy.

Now before you rush to the garden centre and buy a new plant. Wouldn’t it be better if you thought about how to save your plant or prevent it from going without water for an extended period of time?

You could invest in a complex soil moisture sensor system connected to an app that reminds you to water the plant when the soil gets too dry. Or better yet, connect the moisture sensor to an automated watering system. Sounds great, right? But what’s the alternative? You could simply set a calendar reminder to water it every week. Behind both of these solutions lies a simple objective: Make sure the plant doesn’t die.

Whether you’re dealing with day-to-day problems or big data problems, complexity isn’t always the answer.

In the world of data science, there’s a temptation to make things more complicated than they need to be. But as any seasoned gardener—or data scientist—will tell you, simplicity often trumps complexity. In this article, we’ll explore the most common pitfall that teams and individuals frequently encounter when venturing into the realm of data science.

The Lure of Complexity: When Simple Solutions Suffice

Complexity has its allure, especially when you’re trying to impress. However, the principle of Occam’s Razor reminds us that: the simplest solution is often the best one. It’s critical to assess the problem at hand and ask whether a complex model is genuinely necessary or if a simpler approach can do the job effectively.

occam'sgit  razor joke

Machine Learning Isn’t the Be-All and End-All

Machine learning might be the buzzword that captures everyone’s attention, but it’s far from being the only tool in a data scientist’s toolbox. The technology is potent, no argument there, but an over-reliance on machine learning can actually constrain your problem-solving repertoire. Think about it this way: if you’re a chef who only knows how to fry food, you’re missing out on the vast culinary landscape that includes grilling, boiling, and sautéing.

Sometimes, a rule-based system is the most straightforward and effective way to implement business logic. Statistical methods, from chi-squared tests to t-tests, can also offer critical insights into data trends and relationships.

It’s easy to jump on the machine learning bandwagon given its current hype, but let’s not forget the time-tested techniques that laid the foundation of data science as we know it. A simple linear regression model can often yield insights that are just as valuable, if not more so, than a complicated neural network.

“An IF statement in production is worth more than a deep neural network on your laptop” - Naude Pretorius, Head of Internal Modelling at FNB

The secret sauce to successful data science is selecting the most appropriate method for the task at hand. Would you use a sledgehammer to crack a nut? Similarly, deploying a deep learning model for a problem that could be solved with a basic statistical test is overkill. And it’s not just about computational efficiency; it’s about interpretability, ease of deployment, and, most importantly, aligning closely with the problem’s specific needs.


Like a thriving houseplant that, at its core, only requires a simple watering schedule, successful data science projects don’t always need the most complex solutions. Whether you’re new to the field or an experienced practitioner, being aware of this common pitfall can make your journey through the data landscape both successful and fulfilling.

So next time you’re looking for a solution to a data science problem, Ask the question:

“Is this the simplest way to keep the plant alive?”