Two modelsone answer
A Random Forest and a multilayer perceptron, combined into one ensemble and trained on 70,000 patient records. I took it from raw data through cleaning, modelling and evaluation to a peer-reviewed paper at ACET 2025.
Read this firstThe point isn't one clever model. It's carrying a project all the way to a result other researchers checked.
Records split, two models, one answer
Every record goes down both branches at once; the ensemble merges the two outputs into one prediction. Click a stage to read what happens there.
Two kinds of model, one prediction
Every record goes to both models. A Random Forest votes with hundreds of decision trees; an MLP learns smooth patterns across all the features at once. Their outputs are combined into one prediction.
The ideaTrees and neural networks tend to make different mistakes, so combining them can cancel some of each out.
DraftAdd how the outputs were combined and each model's score on its own, to show what the ensemble added.
Out of every 1,000 test cases
Accuracy is easy to say and hard to picture. Here it is as 1,000 test cases: about 948 classified correctly and 52 missed. Hover any dot.
Honest readingAccuracy alone hides which mistakes those 52 are. In a medical setting, a missed case and a false alarm cost very different things.
DraftAdd precision, recall and the confusion matrix from the paper.
Six stages, end to end
The same six stages as the homepage. Here the last stage is a published paper instead of an app.
Why it mattersPeer review means someone outside the project checked the method, not just the number.