Build Better ML Models by Understanding Behavior

In partnership with

Welcome to Gratitude Driven, a weekly newsletter where I share practical ideas and insights across personal growth, professional development, and the world of AI and data science.

Table of Contents

• Better ML By Understanding Humans

• This Book Changed the Way I Think

Better ML By Understanding Humans

I come from a non-technical social science background. While this has meant I’ve needed to do a lot of additional study to strengthen my technical skills, it has also given me some unexpected advantages.

One of those is a strong awareness that the vast majority of machine learning problems are directly influenced by human behavior. This human element shows up everywhere in ML, but it's often overlooked in favor of purely technical approaches and metrics.

Let me give you an example: When I'm working on feature engineering, I've learned that understanding how people actually behave – rather than how we assume they should behave – reveals the patterns that really matter.

Think about building a recommendation system. You could create something that looks perfect on paper, with impressive metrics. But if you haven't considered how people browse when they're tired after work, or how they shop differently on weekends, it won't deliver the value that it could.

Or, imagine you’re creating a model to predict which new creator will be successful on YouTube (I’ve actually done something similar in the past). Sure, we have data on their behavior, but not on the intangibles that really matter, like motivation, drive to learn, and sense of humor. Sometimes this means we will never be able to fully model the complexity — and sometimes it means we need to get really clever about thinking about features that get us as close as possible.

The same principle applies to evaluating model performance. Traditional metrics might tell you your model is performing well, but understanding human context reveals the edge cases that matter most. A chatbot might have high accuracy, but if it doesn't recognize when a user is frustrated or confused, it's not truly effective.

What I've found most valuable from my social science background is the ability to dig into the "why" behind user actions, and think creatively about how we can incorporate human-centric nuance into numerical features. When we understand people's underlying motivations, fears, and aspirations, we can build ML systems that are more performant, effective, and hopefully, useful.

This Book Changed the Way I Think

Last week I read Sebastian Junger’s book Tribe: On Homecoming and Belonging. I thoroughly enjoyed the read, and zoomed through it in a day or two.

The book argues that humans have a deep evolutionary drive to belong to small, purposeful groups ("tribes") and that the loss of these tribal connections in modern society contributes to modern psychological struggles. It explores how adversity and hardship can actually strengthen community bonds and provide a sense of meaning.

A central theme is that humans don't suffer from hardship itself, but rather from not feeling necessary to others, suggesting that finding ways to be useful and contribute to a community is crucial for psychological well-being.

Drawing from these insights, I've developed my own perspective on how we can cultivate this sense of necessity in our modern context.

In my opinion, the path to feeling useful starts with deliberately building capabilities that can benefit others - whether that's learning emergency response skills, developing emotional intelligence for supporting friends, or mastering skills you can teach to others. The crucial next step is actively seeking out opportunities to apply these skills through volunteering, mentoring, or stepping up during community challenges.

If you’re interested in more book recommendations (technical, fiction, and non-fiction), check out my favorites here.

Want to chat 1:1? Book time with me here.

Forwarded this email? Sign up here.

Follow me on YouTube | Instagram | LinkedIn

Note: This email may contain affiliate links. If you make a purchase I may make a small commission, at no cost to you. Thank you for your support!