Machine learning capabilities for the classroom.
Explore the model types, training controls and evaluation methods behind ML4Schools classroom labs. Each guide explains what the capability does, what students can investigate and where the evidence stops.
Prepare the data, build the model, evaluate the result.
ML4Schools combines a Data Playground, ML Lab and Inference Lab. Each exposes a different part of the experiment: the evidence being used, the model being built and the judgement made from its predictions.
These guides explain the supported technical capabilities and their limits. The lesson's dataset, model and teaching design determine which controls are available. Some activities supply a trained model; others let students build and train.
Choose the part of the experiment you want to understand.
- Model building and training
- Neural networks for tabular classification and regression; layers, activations, dropout, optimisers and run comparisons. KNN and SVM where a lesson offers them.
- Image classification
- A frozen feature extractor with a trainable head, or a small CNN from scratch. Training cost, image perturbations and the limits of image evaluation.
- Model evaluation and explanations
- Baseline performance, decision thresholds, mistakes, what-if inputs and supported explanations. Distinguish scores, evidence and causal claims.
- Data preparation
- Audit a dataset, explain preprocessing choices and replay a preparation recipe in ML Lab. Consider what a transformation improves and what it can cost.
The experiment, the teaching and the school workflow.
ML4Schools brings prepared Australian-context learning, configurable model experiments, connected preparation and evaluation, and class delivery into one product. The value is the combination your school can teach and review.
Free tools also offer real model training and substantial teaching resources. Compare ML4Schools with free AI teaching resources, including where a free option may be sufficient.
Model behaviour is evidence to interpret.
A model can perform well on one set of examples and fail on another. An impressive training score, a saliency map or a confident prediction is not enough to establish that the model is suitable for a new purpose.
The technical guides connect the controls students can manipulate with the questions they should ask: which data was used, which model made the prediction, which errors matter and what remains untested.
For concrete classroom contexts, explore Tooth Detective, The Body Atlas and the learning catalogue. For data handling, see student privacy and Security & Trust.