Machine Learning and Predictive Modeling

Active learning works for machine learning because students need to experience the gap between human intuition and algorithmic reasoning firsthand. When they train models themselves, students confront misconceptions immediately through concrete results, not abstract explanations.

• Year 10
• Technologies
• 4 activities
• 25 min–45 min

Learning Objectives

1. Explain how algorithms learn patterns from labeled data in supervised machine learning.
2. Analyze the impact of biased training data on algorithmic fairness and outcomes.
3. Evaluate the limitations of using historical data for predicting future events.
4. Critique the ethical considerations of algorithmic decision-making without human intervention.
5. Design a simple predictive model using a tool like Teachable Machine to classify data.

Simulation Game

⏱ 30 min·Pairs

Pairs Activity: Bias Detection in Datasets

Provide pairs with a sample hiring dataset showing gender bias in promotions. Students chart patterns, calculate error rates for subgroups, and propose balanced alternatives. Pairs present findings to the class.

Prepare & details

How does biased training data lead to discriminatory algorithmic outcomes?

Facilitation Tip: During the Pair Activity on Bias Detection, assign each pair one dataset variant so they compare outputs and argue which skew causes the most unfair classifications.

Setup: Flexible space for group stations

Materials: Role cards with goals/resources, Game currency or tokens, Round tracker

• Apply
• Analyze
• Evaluate
• Create
• Social Awareness
• Decision-Making

Simulation Game

⏱ 45 min·Small Groups

Small Groups: Train Your Own Classifier

Using Teachable Machine or Scratch extensions, groups gather images or text data for categories like fruits or sentiments. They train models, test on new data, and log accuracy drops from poor training sets. Groups swap models to evaluate.

Prepare & details

What are the limitations of using historical data to predict future behavior?

Facilitation Tip: When students Train Your Own Classifier in small groups, circulate and ask: ‘What would happen if this feature disappeared? How would your prediction change?’ to push critical analysis.

Setup: Flexible space for group stations

Materials: Role cards with goals/resources, Game currency or tokens, Round tracker

• Apply
• Analyze
• Evaluate
• Create
• Social Awareness
• Decision-Making

Simulation Game

⏱ 35 min·Whole Class

Whole Class: Ethical Algorithm Debate

Divide class into teams to argue for or against 'Algorithms can make ethical decisions alone.' Prep with case studies like facial recognition errors, then debate with evidence from prior activities. Vote and reflect.

Prepare & details

Is an algorithm capable of making an ethical decision without human intervention?

Facilitation Tip: For the Ethical Algorithm Debate, give students exactly 60 seconds to prepare their opening point after the prompt is revealed to sharpen concise reasoning under time pressure.

Setup: Flexible space for group stations

Materials: Role cards with goals/resources, Game currency or tokens, Round tracker

• Apply
• Analyze
• Evaluate
• Create
• Social Awareness
• Decision-Making

Simulation Game

⏱ 25 min·Individual

Individual: Prediction Journal

Students select a personal dataset like sports scores, build a simple prediction rule, test on new data, and journal limitations like overfitting. Share key insights in a class gallery walk.

Prepare & details

How does biased training data lead to discriminatory algorithmic outcomes?

Facilitation Tip: In the Prediction Journal, require a visual sketch of one failed prediction to make abstract errors visible and discussable.

Setup: Flexible space for group stations

Materials: Role cards with goals/resources, Game currency or tokens, Round tracker

• Apply
• Analyze
• Evaluate
• Create
• Social Awareness
• Decision-Making

Teaching This Topic

Teach this topic by making students confront model failures early and often. Avoid starting with definitions of supervised learning; instead, let them experience training data first. Research shows that confronting misconceptions through active testing embeds durable understanding better than lectures. Focus on small, iterative steps: collect data, train, test, reflect, then revise.

What to Expect

Successful learning looks like students recognizing the limits of raw data, questioning model outcomes, and proposing ethical fixes. They articulate why bias persists, how predictions fail, and what fairness requires beyond technical accuracy.

Watch Out for These Misconceptions

Common Misconception

During the Train Your Own Classifier activity, watch for students assuming the algorithm understands meaning like a human does.

What to Teach Instead

As groups train models, have them intentionally add nonsense features (e.g., ‘number of vowels in email’) and observe how the model still uses them. Then prompt: ‘Why did the model treat this as important? What does this show about how algorithms learn?’

Common Misconception

During the Bias Detection in Datasets activity, watch for students believing that neutral data produces neutral models.

What to Teach Instead

Pairs analyze a loan dataset with missing demographic labels filled in. They recalculate predictions after balancing gender representation, then present how fairness scores changed. Ask: ‘Did the data alone fix the problem? What else matters?’

Common Misconception

During the Prediction Journal activity, watch for students assuming more data always improves predictions.

What to Teach Instead

Students select one prediction (e.g., house price) and graph accuracy versus dataset size. They must explain why accuracy drops when noisy data is added, connecting to real limits like historical redlining data.

Assessment Ideas

Discussion Prompt

After the Ethical Algorithm Debate, pose: ‘Imagine a hiring algorithm that favors male candidates. What steps could a developer take to correct this during Train Your Own Classifier?’ Use student responses to assess their understanding of bias in training data and model evaluation.

Exit Ticket

After the Prediction Journal, ask students to write one prediction an algorithm might make and one limitation or ethical concern. Collect and review for patterns in how students identify data quality, fairness, or model limits.

Quick Check

During the Bias Detection in Datasets activity, present two short dataset descriptions. Ask students to identify which is more likely to lead to a fair model and explain why, using fairness metrics from their earlier analysis.

Extensions & Scaffolding

• Challenge: Ask students who finish early to redesign a dataset so the classifier performs equally well on historically underrepresented groups.
• Scaffolding: Provide a template for the Prediction Journal with sentence starters like ‘The model predicted ____ because ____ but failed when ____.’
• Deeper exploration: Invite students to interview a local professional using predictive modeling (e.g., healthcare, marketing) and report on real-world bias mitigation strategies.

Key Vocabulary

Algorithm	A set of rules or instructions followed by a computer to solve a problem or perform a task, such as making a prediction.
Training Data	The dataset used to teach a machine learning model. The quality and characteristics of this data directly influence the model's performance and fairness.
Supervised Learning	A type of machine learning where the algorithm learns from a dataset that includes both input features and corresponding correct outputs (labels).
Prediction	An output generated by a machine learning model based on patterns learned from data, forecasting a future outcome or classifying an input.
Bias (Algorithmic)	Systematic and repeatable errors in a computer system that create unfair outcomes, often stemming from biased training data or flawed algorithm design.