Z-Score
Z-scores standardise values around a mean, helping analysts compare performance, detect outliers, and benchmark model outputs.
A statistical measure that indicates how many standard deviations a data point is from the mean of a dataset.
What Is a Z-Score
A z-score standardises a value by expressing its distance from the dataset mean in units of standard deviation. A score of 0 means the value is exactly at the mean, positive values are above the mean, and negative values are below it.
How Z-Scores Are Calculated
The formula is: z = (x - μ) / σ, where x is the observed value, μ is the mean, and σ is the standard deviation. This transformation makes values from different scales directly comparable.
Why Z-Scores Are Important in Data Analysis
Z-scores help analysts detect outliers, compare distributions, and normalise features before modelling. They are useful for identifying unusual behaviour in datasets and for creating consistent statistical thresholds across experiments.
Applications in Machine Learning Evaluation
In machine learning, z-scores are used for feature scaling, anomaly detection, and performance interpretation across different metrics. They support more robust comparisons when model outputs have different units or ranges.
Use Cases in AI Research and Model Benchmarking
AI researchers use z-scores to compare model runs, evaluate deviations from baseline systems, and identify statistically significant shifts in quality indicators such as latency, error rates, or translation adequacy scores.
Related Terms
Zero Data Retention
A mode in which user data is deleted almost immediately after processing.
Zero-Shot Learning
A machine learning capability that allows models to perform tasks they were not explicitly trained on by leveraging generalised knowledge.
Zero-Shot Translation
A multilingual translation capability where models translate between language pairs not directly present in training data.
Related Resources
Neural Network
A computational model made up of interconnected layers that learns patterns from data and powers many modern AI systems.
Algorithmic Bias
Systematic errors in AI outputs arising from skewed, imbalanced, or prejudiced training data.
Gender Bias in AI
Differences in how AI treats or represents genders due to training-data patterns.
Jaccard Similarity
A statistical measure used to evaluate the similarity between two sets, often applied in text analysis and information retrieval.
Machine Learning
A field of AI where systems learn patterns from data to make predictions or generate content.
Weight Parameters (Model Weights)
The numerical parameters inside a neural network that determine how input data is processed and how predictions or generated text are produced.