To use or not to use ML. That is the question.

These days, it's hard to go through the news, browse LinkedIn, or engage in conversations with friends without the topic of ChatGPT or artificial intelligence (AI) coming up. These discussions often fall into one of two categories: either filled with optimistic excitement or tainted by apprehension about the inevitable world domination foretold by our robot overlords. As ChatGPT—a substantial language model grounded in machine learning—enables search engines and various products to become smarter and more useful, it's an exceedingly thrilling period to contemplate the possibilities of AI and machine learning (a subset of AI).

In my capacity as a consultant providing data visualization and data science services to businesses, I specialize in constructing machine learning models using algorithms like gradient descent and XGBoost (primarily employed for regression and characterization), and sometimes deep learning models such as Recurrent Neural Networks (RNNs, specifically designed for sequential data tasks like natural language processing (NLP), speech recognition, and time-series forecasting). When I share this with clients and potential clients, they get pretty excited, leading to discussions on how we can leverage machine learning in their projects. While I genuinely appreciate this enthusiasm, and it frequently leads to productive collaborations, the notion of "let's find a use case for machine learning" is akin to putting the cart before the horse or, as exemplified by the meme in this post, reminiscent of "if all you have is a hammer, everything looks like a nail."

Given the current buzz around machine learning, many companies are eager to incorporate this powerful tool into their operations. However, it's crucial to recognize that machine learning is not a universal solution, and there are specific scenarios where it's the appropriate choice. Understanding when it's the right fit and when traditional analytics might be more suitable is essential. Below are some of my recommendations.

Machine learning is best suited for tackling intricate challenges that traditional analytics may struggle to address. For instance, when confronted with extensive and diverse datasets, machine learning algorithms excel at uncovering concealed patterns and trends, delivering invaluable insights to businesses. It excels in predictive modeling, making it ideal for tasks like constructing recommendation systems, detecting fraudulent activities, segmenting customers, and analyzing sentiments in the context of text or language analysis. If your project demands precise predictions and the ability to spot anomalies, machine learning can be a game-changer, enabling you to make data-driven decisions with confidence.

So when should you avoid using machine learning? In cases where the available data is limited, noisy, or lacking in diversity, traditional analytics methods can still yield meaningful results without introducing unnecessary complexity. Sometimes, historical data alone may suffice to draw insightful conclusions, rendering machine learning unnecessary. For more straightforward business problems that do not necessitate forecasting future outcomes, traditional statistical analyses or rule-based approaches may prove more suitable. By comprehending the limitations of machine learning and its reliance on data quality and complexity, you can sidestep unnecessary expenses and ensure your projects yield the maximum value.

I’ll leave you with a critical piece of advice, stemming from my experience in data visualization: Before embarking on any analytical journey, you need to understand your data. That's why, when I collaborate with clients, I typically start with data visualization, followed by a discussion about whether further analysis is warranted and, if so, which approach is most optimal. Seeking to employ analytical tools or programming to address your business inquiries before assessing your specific needs and objectives is akin to waving a hammer around in the dark, which can be dangerous. In conclusion, opt for machine learning when tackling complex challenges, dealing with vast and diverse datasets, and aiming for predictive insights; and use traditional analytics methods for simpler problems, scenarios involving limited data, and historical analyses. By choosing the right tools for the right tasks, you can tailor solutions that effectively address your questions, enabling your business to thrive in a competitive landscape.