04 / Evaluation

Measure what matters.

Evaluation is framed around useful separation, calibration, and failure analysis—not a single headline number.

Interactive evaluation lab

Read the model under pressure.

Move the operating point, inspect class behavior, and compare how the model generalizes. Every chart below uses illustrative demo data.

DEMO DATA

Model

Headline metric

Accuracy87.73%

CNN + BiLSTM

87.7

EfficientNetB0

84.2

MFCC baseline

76.4

Confusion matrix · threshold linked

Pred +Pred −True +8515True −1486

Normalized illustrative counts

ROC curvethreshold marker follows the slider

Training / validation losseight illustrative epochs

stable

Class-wise performancemacro F1 by sound type

90.3%

Decision threshold0.50

Sensitivity 89%Specificity 86%Precision 83%F1 86%

Inspect class coverage before interpreting a result.

See which respiratory sounds are most often confused.

Compare confidence with correctness before review.

Return to the recording and understand the model’s miss.

Prototype statusEvaluation scaffolding is ready for validated study data.