Evaluation, Robustness & Advanced ML

Classification Metrics Simulator

Manipulate TP, TN, FP and FN directly and watch every confusion-derived metric update.

20 · Metrics 21 · Thresholds 22 · Explain 23 · Robustness 24 · Shift 25 · Time series

Lab concept guide

What to observe while you experiment

Classification metrics are different summaries of the same four confusion-matrix counts. Manipulating TP, TN, FP and FN directly reveals the denominators behind accuracy, precision, recall, specificity and F-scores.

MechanismChange one confusion count at a time and identify exactly which metric numerators or denominators are affected.

Failure modeUsing a metric name without defining the positive class or understanding the counts that generate it.

VerificationChoose a small TP/TN/FP/FN configuration and compute at least two metrics by hand before comparing them with the simulator.

Experiment deliberately

Hold TP and FN fixed while increasing FP. Predict which metrics must change and which should remain unchanged, then verify.

Start from counts. Classification metrics are different summaries of the same confusion matrix. Change one count and inspect which metrics react.

True positives42False positives10True negatives38False negatives10

Ready.

Confusion matrix

Rows are actual class; columns are predicted class.

Metric profile

All values are computed from the four cells at left.