TASC Data Science and Digital Solutions key terms

A modeling approach where an algorithm learns from labeled examples, mapping inputs to known outputs so it can predict the target for new, unseen data.

A modeling approach that finds structure in unlabeled data, such as groups or compressed representations, without any target variable to predict.

A supervised task that assigns each observation to one of a set of discrete categories, such as fraud or not fraud, based on its features.

A supervised task that predicts a continuous numeric value, such as a price or a temperature, from one or more input features.

An unsupervised technique that groups observations so that items within a group are more similar to each other than to items in other groups.

An individual measurable input variable used by a model, such as age, income, or a word count, that helps explain or predict the target.

The known target value or correct answer attached to a training example, which a supervised model learns to predict from the features.

The portion of data used to fit a model's parameters, kept separate from evaluation data so performance estimates remain honest.

A held-out portion of data never used during model fitting, reserved for estimating how well the final model generalizes to new cases.

A resampling method that splits data into several folds, training and validating repeatedly, to estimate performance more reliably than a single split.

A failure mode where a model learns noise and quirks of the training data, performing very well there but poorly on new, unseen data.

A technique that penalizes model complexity, for example by shrinking large coefficients, to reduce overfitting and improve generalization to new data.

A table comparing predicted versus actual classes, showing counts of true positives, true negatives, false positives, and false negatives for a classifier.

A classification metric giving the share of predicted positives that are actually positive; high precision means few false alarms.

A classification metric giving the share of actual positives that the model correctly identified; high recall means few missed cases.

The practice of creating, transforming, or selecting input variables, such as ratios or date parts, so a model can learn patterns more effectively.

The process of detecting and correcting errors in data, including duplicates, inconsistent formats, impossible values, and missing entries, before analysis.

Absent values in a dataset, which analysts handle by removing records, imputing estimates, or using models that tolerate gaps, depending on the cause.

Replacing a missing value with an estimated substitute, such as a mean, median, or model-based prediction, so complete-case analysis becomes possible.

An observation that lies unusually far from the rest of the data, which may signal a data error or a genuinely rare and important event.

An initial, open-ended pass through a dataset using summaries and charts to reveal distributions, relationships, anomalies, and questions worth deeper investigation.

Summary measures that characterize a dataset, such as mean, median, standard deviation, and range, used to describe central tendency and spread.

The pattern describing how often each value or range of values occurs in a variable, including its shape, center, and spread.

A symmetric, bell-shaped probability distribution defined by its mean and standard deviation, common in theory and often assumed by statistical tests.

A measure of how spread out values are around the mean, equal to the square root of the variance and expressed in the data's original units.

A statistic between minus one and one describing the strength and direction of a linear relationship between two variables; it does not prove causation.

A relationship where changing one factor actually produces a change in another, which requires stronger evidence than an observed association or correlation.

A formal procedure comparing observed data against a null hypothesis, producing a p-value that indicates how compatible the data are with that null.

The probability of observing results at least as extreme as the data, assuming the null hypothesis is true; it is not the probability the null is true.

A conclusion that an observed result is unlikely under the null hypothesis given a chosen threshold, indicating an effect probably exists but not how large it is.

A controlled experiment comparing two versions of something, randomly assigning users to variants and measuring a chosen outcome to judge which performs better.

An automated sequence of steps that moves data from sources through transformation and validation into storage or a model, run reliably and repeatedly.

Extract, transform, load: a pipeline pattern that pulls data from sources, cleans and reshapes it, then writes it into a target store such as a warehouse.

A central repository that consolidates cleaned, structured data from many sources, optimized for querying and business reporting rather than transaction processing.

A storage system holding large volumes of raw data in native formats, structured or not, with structure applied at read time rather than write time.

The policies, roles, and controls that define who may access data, how quality is maintained, and how privacy and regulatory obligations are met.

The degree to which data is accurate, complete, consistent, timely, and fit for its intended use in analysis or operations.

The protection of personal information through practices such as minimizing collection, restricting access, and honoring consent and legal requirements.

Removing or altering identifying details so individuals cannot reasonably be re-identified, while preserving the data's analytical usefulness.