Underfitting
Underfitting is the failure mode teams under-diagnose: the model is too simple (or under-trained, or starved of signal) to even fit its own training data. The tell is a poor TRAINING score with a small train/validation gap, and flat parallel learning curves. The fixes go the opposite way from overfitting: add features, capacity, and training — remove regularization — in that order.
01.The Problem: The Metric That Will Not Move, No Matter How Many Labels You Buy
Here is a story that plays out in real companies constantly.
A team ships a first model. Accuracy: 0.70. Management says "data is the answer" and funds a labeling campaign. Ten thousand more examples later:
Accuracy: 0.701.
Someone proposes more regularization. Accuracy: 0.698. Someone else proposes an ensemble of the same model. 0.70.
The team concludes "the problem must be inherently hard" and moves on.
Meanwhile the actual issue was staring at them the whole time: the model could not even fit the data it was trained on. Training accuracy was also 0.70. Same as validation. The gap was tiny.
That is the opposite failure to overfitting (Topic 52: a model memorizes its training set and fails on new data). There, the model is too flexible. Here, the model is too stiff — like trying to trace a curvy coastline using only a straight ruler. No number of extra maps will make the ruler draw curves.
So the question becomes
How do we tell "the model is too simple / badly trained" apart from "the problem is impossible" — and what do we do about it?
Most textbooks spend one paragraph on this failure. In practice, underfitting is more common than the textbooks suggest, because real projects are limited by feature quality and training budget far more often than by model flexibility. This topic gives you the signature, six separable causes, and the fix order.
Underfitting Branch of the Diagnosis Tree 📉
Underfitting Branch of the Diagnosis Tree 📉
The gap is small but the level is low — that combination means the hypothesis class, the features, or the training budget are limiting, and regularization is the wrong lever.
Unlock Topic #53: Underfitting
You are viewing a preview. The full in-depth technical walkthrough, worked derivations, and code notebooks for this concept, along with self-assessment quizzes, are available with Pro or Lifetime Access.
Failure modes, high-throughput bottlenecks, and real FAANG implementation decisions.
Interactive system topology diagrams, live parameter simulators, and downloadable SVG charts.
Staff-level multiple-choice quiz questions with instant feedback and answer explanations.
Firebase Google authentication automatically syncs your completed topics and quiz scores.
How clear and actionable was this distributed systems breakdown?