When a model is too simple, it misses the real pattern
🔬
Professor Spark
Underfitting is the opposite of overfitting. The model is so simple it cannot capture the true relationship — even on the data it has already seen. Your task: recognise it and fix it.
Three missions
You will face three situations. In each one you must decide whether the current model is underfitting, and then improve it if needed. Observation and reasoning required — no free points.
Mission 1 – Spot the underfit
Look at the data and the current model (the flat line). Is this model underfitting?
–
Current MSE
Very high
Error level
Yes – the model is too simple and ignores the clear upward trend
No – this is a good model; the data is just random
No – this is overfitting because the line is perfect
Mission 2 – Choose the right complexity
The data has a clear rising pattern. Three candidate models are offered. Which one is the sensible choice (neither under- nor over-fitting)?
Model A – flat horizontal line (ignores hours completely)
Model B – straight line with positive slope
Model C – extremely wiggly curve that hits every point
Mission 3 – Fix the underfit yourself
The current line is almost flat (underfitting). Raise the slope until the MSE drops below the target. You must understand what “more complexity” means here: giving the model the ability to follow the trend.
–
Your MSE
15
Target (max)
Challenge Complete
You now know the two classic failures of models:
Underfitting – model too simple, high error even on training data, misses the real pattern.
Overfitting – model too complex, almost zero training error, but fails on new data.
The sweet spot is a model that is complex enough to capture the true relationship, yet simple enough to ignore noise.