35.6 Linear Model Evaluation and Prediction

Linear Model Evaluation and Prediction is the process of using a constructed linear model to calculate a specific predicted output for a chosen input, then interpreting that prediction responsibly with attention to its units, its relationship to already-known data, and the practical boundaries of the situation the model describes. Building directly on numerical function evaluation, this process applies the same substitution and simplification steps to a contextual linear model specifically, adding the additional interpretive care required when a prediction is meant to describe something in the real world rather than an abstract number alone.

Because a linear model extends its constant rate of change beyond whatever specific data originally informed it, this process pays particular attention to distinguishing predictions made within the range of already-known data from those extending beyond it, since the reliability of a prediction can depend heavily on which of these two situations applies.

Allowed Input Selection

Choosing an Input Appropriate to the Question Being Asked

The specific input value used for a prediction is chosen based on the particular question the model is being asked to answer, such as selecting a specific number of hours to predict a corresponding total cost.

Confirming the Chosen Input Respects Any Contextual Restriction

Before proceeding, the chosen input is checked against any contextual domain restriction established during the model's original construction, confirming that the selected value is actually meaningful within the situation the model describes.

Selecting Multiple Inputs for a Series of Predictions

Where more than one prediction is needed, a series of input values can be selected in advance, allowing a sequence of predictions to be calculated efficiently using the same constructed model.

Input Substitution into the Model

Substituting the Chosen Input

The selected input value is substituted into the linear model's rule at every occurrence of the independent variable, following the same replacement completeness standard established under replacement of every input occurrence.

Grouping Negative or Fractional Inputs

Where the chosen input is negative or fractional, it is enclosed in its own parentheses during substitution, following the same negative input grouping and fractional input grouping practices used in general function evaluation.

Preparing the Substituted Expression for Simplification

Once substitution is complete, the resulting expression contains only numbers, ready to be simplified down to a single predicted value in the step that follows.

Predicted Output Calculation

Simplifying the Substituted Expression

The substituted expression is simplified using the standard order of operations, following the same post-substitution operation order used throughout numerical function evaluation, producing a single numerical result.

Confirming the Calculation Through Rechecking

Because a prediction is often used to inform a real decision or answer a specific question, rechecking the arithmetic independently, following the same spirit as function rule arithmetic recheck, provides added confidence in the calculated result.

An Example of a Predicted Output Calculation

Using the model $$f\left(x\right)=5x+20$$, the predicted output at an input of $$6$$ is calculated as $$f\left(6\right)=5\left(6\right)+20=50$$.

Prediction Unit Attachment

Attaching the Correct Unit to the Result

Once calculated, the predicted output is reported together with its correct unit, matching the unit established during model unit interpretation when the model was originally constructed.

Why Units Cannot Be Omitted From a Prediction

A predicted number without its unit conveys substantially less information than the same number with its unit attached, since the bare number alone does not communicate what quantity it actually represents within the modeled situation.

Confirming Consistent Units Throughout the Prediction Process

Before finalizing a prediction, the units of the input and the units of the resulting output are both reviewed together, confirming that they remain consistent with how the model's variables were originally declared.

Known Data Interpolation

Predicting Within the Range of Already-Known Data

Interpolation refers to predicting an output for an input value that falls between two input values already known from the original data used to construct or verify the model.

Why Interpolated Predictions Are Generally More Reliable

Because interpolated predictions fall within the range where the model's behavior has already been directly observed or confirmed, they generally carry a higher degree of reliability than predictions made outside that already-confirmed range.

An Example of Interpolation

If a model was constructed from data at inputs of $$2$$ and $$8$$, a prediction made at an input of $$5$$ falls between these two known values and is therefore an interpolated prediction.

Limited Extrapolation

Predicting Beyond the Range of Already-Known Data

Extrapolation refers to predicting an output for an input value that falls outside the range of inputs already known from the original data, relying entirely on the assumption that the same constant rate continues to apply beyond that observed range.

Why Extrapolated Predictions Carry Greater Uncertainty

Because extrapolation assumes the modeled relationship remains linear beyond the region where it was actually confirmed, extrapolated predictions carry inherently greater uncertainty than interpolated ones, particularly as the chosen input moves further away from the known data.

Using Extrapolation Cautiously

Extrapolated predictions are best treated as reasonable estimates based on the assumption of continued linear behavior, rather than as guaranteed outcomes, with appropriate caution increasing as the extrapolated input moves farther from the range of originally known data.

Context Boundary Check

Checking Whether a Prediction Falls Within a Contextual Boundary

Beyond checking whether an input is mathematically allowed, the resulting predicted output is checked against any practical boundary implied by the situation being modeled, such as confirming that a predicted remaining quantity does not fall below zero when the situation would not allow a negative amount.

Recognizing When a Model Breaks Down at a Boundary

If a prediction violates a known practical boundary, this signals that the linear model may no longer accurately describe the situation at that particular input, even if the underlying arithmetic was performed correctly.

Responding to a Detected Boundary Violation

When a boundary violation is detected, the specific input responsible is treated as falling outside the model's contextually meaningful domain, following the same domain restriction principle established under contextual domain restriction, rather than accepting the resulting prediction at face value.

Model Prediction Interpretation

Stating the Prediction in Context

A complete interpretation restates the calculated prediction using the specific language and units of the original situation, connecting the numerical result directly back to the question that motivated the prediction in the first place.

Distinguishing a Prediction From a Guaranteed Outcome

A model's prediction represents what the linear relationship implies given its constant rate assumption, not necessarily a guaranteed real-world outcome, particularly for extrapolated predictions where the underlying assumption of continued linear behavior is least directly supported by observed data.

Communicating Confidence Appropriately

A well-communicated prediction distinguishes between interpolated results, generally presented with greater confidence, and extrapolated results, generally presented with appropriate qualification about the additional uncertainty involved in assuming the model's constant rate continues to hold beyond the originally known data.