Accuracy

Accuracy is the percentage of predictions a model gets right across all predictions made. A model that correctly identifies 99 out of 100 items scores 99% accurate — a number that sounds impressive until you consider what it might be missing. In problems where one outcome is rare, a model can achieve high accuracy by simply predicting the common outcome every time, never catching the actual cases of interest. That makes accuracy a useful starting point but a poor final answer, especially when the cost of different types of errors varies.

Decisions get made from whatever metric gets reported. A team that surfaces 99% accuracy is answering the question "how often is it right overall?" — when the real question is often "how often does it catch fraud, flag the defect, or identify the at-risk patient?" Those require precision, recall, and error-cost analysis. Approving a model on accuracy alone, without asking what it's missing, is how organizations deploy systems that look good on paper and fail in practice.