Intro to Machine Learning for Teens (14–17): 5 Sessions Using Spreadsheets First

Why start machine learning with spreadsheets (not code)?

If your teen is 14–17 and curious about AI, it’s tempting to jump straight into Python notebooks and complex libraries. But for a true machine learning for teens beginner experience, spreadsheets are the fastest way to build “data intuition” without getting stuck on setup, syntax, or debugging.

Spreadsheets let teens focus on the core ML ideas:

• What is a dataset? Rows, columns, features, labels.
• What patterns exist? Trends, clusters, outliers.
• What makes a model “good”? Accuracy, error, and fairness.
• How do predictions happen? Inputs → rules/weights → outputs.

Parents like this approach because it’s concrete and measurable. Teens like it because they can move quickly and see results in minutes.

This post gives you an intro to ML high school lesson plan you can run at home or adapt for a classroom—five sessions, real data, spreadsheets first, and coding optional at the end.

---

What you’ll need (and how to pick real data safely)

Tools (free):

• Google Sheets (recommended) or Excel
• A dataset source (below)
• 45–60 minutes per session

Good beginner-friendly data sources:

• Google Trends (exportable summaries)
• NOAA climate data (temperature/rainfall)
• City open data portals (traffic counts, park usage)
• Kaggle datasets (choose “small and clean”)

A safe, teen-appropriate dataset checklist:

• Avoid personal identifiers (names, emails, addresses)
• Keep it under ~5,000 rows for smooth spreadsheet performance
• Include at least one column that can be a target (something to predict)
• Include at least 3–6 meaningful features (inputs that might help)

To make this truly “real data” without privacy risks, a great starting point is a public dataset about the world, not people. Think weather, sports stats, movie runtimes and ratings, public transit delays, or school energy usage (anonymized).

---

The 5-session path: from messy data to ML thinking

Each session below is designed for one sitting. If your teen is motivated, they can double up—just keep the reflection questions.

Session 1 (45–60 min): Data detective—cleaning and asking the right question

Goal: Turn a dataset into a clear prediction question.

What to do in Sheets:

• Import the dataset (File → Import).
• Identify:
  • Features: the input columns
  • Label/target: what you want to predict
• Clean the basics:
  • Remove empty rows
  • Standardize categories (e.g., “NYC” vs “New York City”)
  • Convert numbers stored as text
  • Create a “clean” tab so the original stays untouched

Mini-project prompt:

Pick one of these prediction questions:

• Weather: “Can we predict tomorrow’s temperature from today’s?”
• Movies: “Can we predict a movie’s rating from its budget and genre?”
• Transit: “Can we predict if a train will be late based on time of day and station?”

Reflection questions (parents can ask):

• What column is the target, and why?
• Which columns might be cheating (e.g., contain the answer indirectly)?
• What’s one data quality issue that could mislead a model?

---

Session 2 (45–60 min): Patterns first—charts, correlation, and “feature sense”

Goal: Teach that ML starts with patterns, not magic.

What to do in Sheets:

• Make a few quick visuals:
  • Scatter plot (feature vs target)
  • Line chart (time trends)
  • Bar chart (averages by category)
• Add basic summary stats:
  • AVERAGE, MEDIAN, MIN, MAX
• Compute correlation (when numeric):
  • =CORREL(feature_range, target_range)

Teach one key idea: correlation is not causation—but it can guide which features are useful.

Quick win:

Have your teen pick their “Top 3 features” they believe will help prediction and justify them in one sentence each.

---

Session 3 (45–60 min): Build a simple model in a spreadsheet (linear regression)

Goal: Create the first real predictive model using spreadsheet tools.

There are two easy routes:

• Google Sheets trendline (fastest): add a scatter plot, add a trendline, show equation and R².
• LINEST function (more powerful): get coefficients directly.

Option A: Trendline model (beginner-friendly)

1. Make a scatter plot: X = one feature, Y = target.
2. Customize → Series → Trendline → Linear.
3. Turn on “Label: Use Equation” and “Show R²”.

Now your teen has an equation like:

• Target ≈ (m × Feature) + b

Option B: LINEST model (best for multiple features)

Use =LINEST(target_range, feature_matrix, TRUE, TRUE) to estimate coefficients.

Important concept to explain simply:

• The model “learns” weights (coefficients). Bigger weight = bigger influence.

Make predictions in a new column:

• =m*X + b (or sum of weights across multiple features)

---

Session 4 (45–60 min): Evaluate it like a scientist—error, train/test split, and overconfidence

Goal: Teach that ML is about testing, not just building.

In spreadsheets, teens often “accidentally” evaluate on the same data they trained on. This is a perfect moment to introduce a core ML habit: hold-out testing.

Simple train/test split in Sheets:

• Add a random column: =RAND()
• Sort by that column
• Use first 80% of rows as train, last 20% as test

Evaluate with easy metrics:

For numeric prediction (regression):

• Error: =Actual - Predicted
• Absolute error: =ABS(Actual - Predicted)
• Mean Absolute Error (MAE): average of absolute errors

For yes/no prediction (classification) you can do later, but if you do it now:

• Convert predicted score to yes/no using a threshold (e.g., >0.5)
• Accuracy = percent correct

Here’s a practical table you can follow session-by-session:

Session	What your teen builds	Spreadsheet skill	“ML idea” learned	Output to save
1	Clean dataset + prediction question	Import, filters, cleaning tab	Features vs target	A one-sentence ML question
2	Charts + correlations	Scatter plots, CORREL, pivot summary	Patterns and signals	Top 3 candidate features
3	First model	Trendline or LINEST	Weights/coefficients	Predictions column
4	Train/test evaluation	RAND split, error formulas	Generalization	MAE/accuracy on test set
5	Improve + present	Pivot tables, what-if checks	Fairness + iteration	One-slide “model report”

Conversation starter:

• “Does it work equally well on new data?”
• “What kinds of examples does it fail on?”

This is also where teens learn a subtle lesson: a high R² or high accuracy doesn’t automatically mean a model is useful.

---

Session 5 (45–60 min): Improve the model—feature engineering, fairness checks, and a teen-friendly ML report

Goal: Make the project feel real: iterate, check bias, and communicate results.

Part A: Improve with simple feature engineering

Try 1–2 transformations:

• Create a “month” column from a date (seasonality)
• Bucket a numeric feature into ranges (e.g., low/medium/high)
• Add an interaction feature (e.g., feature1 * feature2)
• Normalize a feature (subtract mean, divide by standard deviation)

Part B: Run a fairness / consistency check (simple but meaningful)

Even with non-sensitive data, you can check whether errors are uneven across groups.

Example:

• If predicting transit delays, compare MAE by line or station.
• If predicting grades (use synthetic or anonymized!), compare error by class period or assignment type, not by student.

Use a pivot table:

• Rows: group (e.g., station)
• Values: average absolute error

If one group has much worse errors, discuss why:

• Is the data smaller for that group?
• Are there missing features that matter?
• Is the group systematically different?

Part C: Create a 1-slide ML report

Have your teen produce a short report (Google Slides or a single sheet tab) with:

• The prediction question
• Dataset source and what was cleaned
• Features used (and why)
• Test result (MAE or accuracy)
• 2–3 failure cases and what they suggest
• One ethical note: what the model should not be used for

This presentation step is what turns “I played with data” into “I understand ML.”

---

Common parent questions (and practical answers)

“Does this count as real machine learning if it’s in Sheets?”

Yes. ML is the process—defining a target, learning patterns from data, and testing on new examples. Spreadsheets are just a friendly interface.

“When should my teen start coding?”

After Session 4, when they can explain train/test split and error in plain language. Then coding becomes a tool, not a hurdle.

“My teen is advanced—how do we make it harder?”

• Add multiple features (LINEST with several columns)
• Compare two models (one-feature vs multi-feature)
• Try a simple classification problem with a threshold
• Track results in a model comparison table

“How does this fit ‘data science for teenagers at home’?”

It’s realistic: data cleaning, charting, modeling, evaluation, and communication—exactly what data science looks like in the real world.

---

Next Steps: How to get started this weekend

1. Pick a dataset in 10 minutes
   • Choose something your teen already cares about (sports, climate, music trends, transit).
2. Schedule five sessions on the calendar
   • Even two sessions per weekend works.
3. Use the table above as your checklist
   • Save one “artifact” each session (question, chart, model, metric, report).
4. Add one stretch goal (optional)
   • After Session 5, recreate the same model in code (Python) only if your teen wants to.
5. Make it social
   • Have them teach the project to a friend or family member in 3 minutes. If they can explain it clearly, they understand it.

If you want a guided version with interactive feedback, Intellect Council’s teen pathways build these exact habits—cleaning data, making predictions, and learning how to think critically about results—so your teen doesn’t just “use AI,” they understand it.