Understanding Overfitting and Underfitting in Machine Learning

Overfitting occurs when a model memorises training data, achieving 99 % accuracy on the training set but only 61 % on unseen data, as illustrated in the article. The author explains that overfitting is a form of model complexity that captures noise rather than underlying patterns, while underfitting happens when the model is too simple to capture the true signal. The bias‑variance trade‑off is described without heavy math, showing how increasing model complexity can reduce bias but increase variance. A learning curve is used to detect overfitting by plotting training versus validation error across epochs. Practical fixes such as reducing polynomial degree, adding regularisation, or using cross‑validation are demonstrated with a Python example that trains three polynomial regression models of degrees 1, 3, and 15. The visual comparison shows the underfit, good fit, and overfit models, reinforcing the importance of balancing model complexity for real‑world performance.