What Are Input Variables?

Input variables are the quantities supplied to a system, model, or process that determine or influence its behavior and outputs. In engineering and applied mathematics, they are formally distinguished from output variables, which describe the system's response, and from internal state variables, which capture information the system retains between time steps. The concept appears across control engineering, signal processing, statistics, and machine learning, each domain applying the same basic idea with different vocabulary and formal machinery.

In a dynamical system, the input at each moment determines how the state evolves and what outputs are produced. Identifying which external quantities are inputs, and distinguishing among them by type, is a prerequisite for designing controllers, building predictive models, and diagnosing system behavior. The classification of input variables into manipulated, disturbance, and reference categories is standard in classical control theory and carries into process engineering, where it defines the degrees of freedom available for regulation.

Manipulated Variables

Manipulated variables are inputs that a controller or operator can directly adjust to drive a process toward a desired operating point. In a feedback control loop, the controller computes a corrective action and applies it through the manipulated variable, such as the opening of a valve, the speed of a pump, or the voltage applied to an actuator. The Cornell processdesign resource on process controls defines manipulated variables as quantities directly adjusted to control the system, and illustrates the concept with a surge tank example in which the outlet valve position is the manipulated input and the tank liquid level is the controlled output.

In feedforward control, the manipulated variable is adjusted in anticipation of a measured disturbance before the disturbance has affected the output. This is possible only when the disturbance is measurable and a predictive model of its effect on the output is available. Feedforward and feedback schemes are often combined to exploit the advantages of both.

Disturbance Variables

Disturbance variables are inputs that affect the system output but lie outside the controller's authority. External load changes, ambient temperature shifts, raw material composition variations, and supply pressure fluctuations are typical disturbances in industrial processes. Disturbances may be measurable, in which case they can be incorporated into a feedforward compensation scheme, or unmeasurable, in which case the feedback path must absorb their effect through error correction.

In signal processing, noise is the analogous concept: a stochastic input superimposed on the intended signal that degrades the fidelity of the output. Kalman filtering and robust control design both address the challenge of estimating system states and generating control actions when disturbances are present but imperfectly known. The Caltech feedback fundamentals lecture notes cover the mathematical treatment of disturbance rejection in feedback loops, including sensitivity and complementary sensitivity functions that quantify how effectively a given controller attenuates input disturbances.

Input Variable Selection in Modeling and Machine Learning

When building predictive models from data, choosing which input variables to include is a critical step that affects both accuracy and interpretability. In classical regression, variable selection methods such as stepwise regression and penalized estimation (LASSO, ridge) identify subsets of candidate predictors that explain output variance without overfitting. In machine learning, the equivalent task is feature selection, and methods range from filter approaches based on mutual information or correlation coefficients to wrapper methods that evaluate subsets by their effect on model performance. The IBM overview of feature selection describes how retaining only the most informative input variables reduces computational cost and improves generalization by preventing models from fitting noise in irrelevant predictors.

Applications

Input variables are a foundational concept in a wide range of fields, including:

• Industrial process control and optimization
• Automatic feedback controller design
• Statistical regression and econometric modeling
• Machine learning feature engineering and model training
• Signal processing and filter design
• Simulation and digital twin development