FROM FEATURES TO FINDINGS

Small model. Clear insights.

Explore how nine cell characteristics become a prediction.

Explore a prediction

Loading the saved experiment…

A strong result, with a caveat. Training and testing share — sample IDs. These results are from one fixed split.

Review data quality

LOOK INSIDE THE MODEL

Prediction lab

Interactive · runs on your device

01 Cell characteristics02 Class likelihoods × prior03 Predicted class

Start with a real sample

Real test sample

Feature scores range from 1 to 10. Adjust a value to explore a hypothetical sample.

What is driving this prediction? Inspect feature contributions

Each bar shows the log likelihood ratio for one feature. Left favors benign; right favors malignant. The class prior also contributes to the final score.

THE HELD-OUT RESULTS

Every prediction, accounted for.

Prediction breakdown

Test set

Actual class → predicted class

Malignant is the positive class throughout this dashboard.

Test samples

Sample ID	Actual	Prediction	Result	Explore sample

CONTEXT BEHIND THE NUMBERS

Data & methodology

Source-backed · no demonstration metrics

▤

A fixed experiment

Nine integer features, scored 1–10. Sample IDs and actual labels are excluded from prediction. No missing feature values in the supplied matrices.

SOURCE DATACasosTrainTestWisconsin.matVERIFIED AGAINSTClasificador.mat

Check independence first

shared sample IDs

identical full rows appear in both partitions. Overlap can make evaluation optimistic; repeated IDs also need review.

Recommended next step

Deduplicate records, split by sample ID, then reevaluate before comparing models.

Know what this measures

Categorical Naive BayesFrequency tables with Laplace smoothing (α = 1).
One supplied train/test splitNo repeated runs or external validation are included.
Explicit class conventionThe original script uses benign as positive. Here, malignant is positive.

This is an educational experiment, not a validated diagnostic tool. Model support is not a calibrated clinical risk estimate.