What Is Accuracy?

Accuracy is a property of a measurement, prediction, or system output that describes how closely a result agrees with the true or accepted reference value. In metrology, accuracy is defined qualitatively as the closeness of agreement between a measured quantity value and the true value of the measurand; it is a concept, not a number, and quantitative expressions of closeness belong to measurement uncertainty rather than to accuracy itself. In computing and data science, accuracy carries a more operational definition: the fraction of correct outputs produced by a system under evaluation, expressed as a ratio or percentage over a test set.

The distinction between accuracy and related concepts such as precision, trueness, and error appears throughout engineering and scientific work, and conflating them leads to faulty instrument specifications, misleading classifier evaluations, and incorrect uncertainty budgets.

Accuracy and Precision in Metrology

The international vocabulary of metrology (VIM), maintained jointly by ISO, BIPM, and partner organizations, distinguishes accuracy from precision in a way that parallels the target analogy: a set of measurements is accurate if their average lies near the true value; it is precise if the measurements cluster tightly together regardless of where that cluster falls. Measurement accuracy encompasses both trueness (freedom from systematic bias) and precision (freedom from random scatter). NIST Technical Note 1297, which provides guidelines for evaluating and expressing measurement uncertainty, follows the VIM framework and specifies that accuracy should not be assigned a numerical value; instead, the Guide to the Expression of Uncertainty in Measurement (GUM) defines a structured process for quantifying and combining the uncertainty components that characterize a measurement result. Calibration against traceable reference standards reduces systematic error and improves trueness; reducing random error requires repeated measurement and statistical averaging.

Measurement Accuracy and Uncertainty

Every measurement carries uncertainty arising from the instrument, the measurement procedure, the operator, environmental conditions, and the reference standard used for calibration. Uncertainty is expressed as a range, typically at a stated coverage probability, within which the true value is expected to lie. Instrument specifications use terms such as reading error, full-scale error, linearity error, and hysteresis to characterize accuracy components. In sensor systems and data acquisition hardware, manufacturers specify accuracy as a percentage of full-scale deflection or as an absolute value in engineering units. NIST's guidance on measurement uncertainty provides the statistical framework for combining these components using Type A evaluations (based on statistical analysis of repeated observations) and Type B evaluations (based on other means, such as calibration certificates or manufacturer specifications).

Accuracy in Classification and Computational Systems

In machine learning and pattern recognition, accuracy is defined operationally as the number of correct classifications divided by the total number of test cases. A classifier with 95 percent accuracy correctly labels 19 out of every 20 inputs. This metric, while intuitive, can be misleading when class distributions are unbalanced: a classifier that always predicts the majority class in a dataset where 95 percent of examples belong to that class will achieve 95 percent accuracy while providing no useful discrimination. Google's machine learning classification guides explain why precision, recall, and F1 score complement raw accuracy in evaluating classifiers with imbalanced classes. IEEE Instrumentation and Measurement literature has noted that the overlapping terminology between metrology and machine learning can create confusion, with words like accuracy and precision carrying subtly different meanings in each domain.

Applications

Accuracy has applications in a wide range of disciplines, including:

• Test and measurement instrument specification and calibration
• Medical diagnostic imaging and clinical assay performance validation
• Autonomous system positioning, navigation, and localization
• Machine learning model evaluation and regulatory AI auditing
• Environmental monitoring and sensor network data quality assurance