aimnova.
DashboardMy LearningPaper MasteryStudy Plan

Stay in the loop

Study tips, product updates, and early access to new features.

aimnova.

AI-powered IB study platform with personalised plans, instant feedback, and examiner-style marking.

IB Subjects
  • All IB Subjects
  • IB Diploma
  • IB ESS
  • IB Economics
  • IB Business Management
  • IB Math AI
  • IB Math AA
Question Banks
  • ESS Question Bank
  • Economics Question Bank
  • Business Management Question Bank
  • Math AI Question Bank
  • Math AA Question Bank
Predicted Topics 2026
  • ESS Predictions 2026
  • Economics Predictions 2026
  • Business Management Predictions 2026
  • Math AI Predictions 2026
  • Math AA Predictions 2026

Study Resources

  • Free Study Notes
  • Mock Exams
  • Revision Guide
  • Flashcards
  • Exam Skills
  • Command Terms
  • Past Paper Feedback
  • Grade Calculator
  • Exam Timetable 2026

Company

  • Features
  • Pricing
  • About Us
  • Blog
  • Contact
  • Terms
  • Privacy
  • Cookies

© 2026 Aimnova. All rights reserved.

Made with 💜 for IB students worldwide

v0.1.895
NotesMath AI HLTopic 4.13Non-linear regression
Back to Math AI HL Topics
4.13.12 min read

Non-linear regression

IB Mathematics: Applications and Interpretation • Unit 4

Smart study tools

Turn reading into results

Move beyond passive notes. Answer real exam questions, get AI feedback, and build the skills that earn top marks.

Get Started Free

Contents

  • Curved data → fit a non-linear model
  • R² and choosing the best model
If the scatter bends, a straight line is the wrong tool: Plot the data. If it curves — bacteria multiplying, a hot drink cooling, light fading with distance — a straight line will miss it badly.

Instead pick a model whose shape matches the picture, and let the GDC find its constants. The common AI HL families are:

• Power y = a·xᵇ — growth/decline that starts steep then flattens (areas, biology scaling). • Exponential y = k·aˣ (or k·eʳˣ) — constant % change (growth or decay). • Quadratic / cubic — a rise-then-fall, or an S-shaped wiggle. • Sinusoidal y = a·sin(bx + c) + d — anything that repeats (tides, daylight).

The GDC does the arithmetic; your job is choosing the right family.
On the GDC (allowed on every AI paper): Enter the x-values in one list and the y-values in another, then choose the regression that matches the shape (ExpReg, PwrReg, QuadReg, CubicReg, SinReg…).

The GDC returns the constants (a, b, k, r…) and R² in one go. Write the model with those numbers substituted in.

IB-style question — fit an exponential decay model

A capacitor discharges. The voltage V (volts) is measured t seconds after the switch opens.

t: 0, 1, 2, 3, 4 V: 12.0, 7.8, 5.1, 3.3, 2.2

Fit a model of the form V = k·eʳᵗ and state R².

Step by step

  1. The values fall by roughly the same FACTOR each second (≈0.65×), so this is exponential decay — use ExpReg of the form k·eʳᵗ on the GDC.
  2. Entering the lists and running the regression gives the constants.
  3. Write the model with the numbers in, and read off the goodness-of-fit value.

Final answer

V = 12.0·e−0.425t, with R² = 0.9998 — an excellent fit. (r < 0 confirms decay.)

R² scores the fit between 0 and 1: The coefficient of determination R² measures how much of the variation in the data the model explains.

• R² = 1 → the curve passes through every point (perfect fit). • R² close to 1 (say 0.95+) → a strong fit. • R² near 0 → the model barely beats just guessing the mean.

To choose between two candidate models, fit both on the GDC and compare their R² — the higher one usually wins. But also sanity-check the shape and whether it stays sensible in context (e.g. a population can't go negative).
SSres = sum of squared residuals (data − model). A smaller SSres means a higher R² and a better fit.

IB-style question — compute a sum of squared residuals

A quadratic model gives predicted heights ŷ for three data points:

x: 1, 2, 3 y (data): 4.0, 9.0, 16.0 ŷ (model): 4.2, 8.5, 16.1

Find the sum of squared residuals, SSres.

Step by step

  1. A residual is data − model for each point.
  2. Square each residual and add.
  3. Total the squares.

Final answer

SSres = 0.30 volts² (smaller is better). Squaring stops positive and negative residuals from cancelling.

IB-style question — pick the better model

Cooling-coffee data is fitted two ways. The GDC reports R² = 0.982 for a power model and R² = 0.997 for an exponential model.

State, with a reason, which model fits better.

Step by step

  1. Compare the two R² values — the larger value explains more of the variation.
  2. Interpret: the exponential model leaves smaller residuals.

Final answer

The exponential model fits better because its R² (0.997) is closer to 1 than the power model's (0.982).

Try an IB Exam Question — Free AI Feedback

Test yourself on Non-linear regression. Write your answer and get instant AI feedback — just like a real IB examiner.

A model predicts heights ŷ = 3.1, 6.0, 9.2, 11.8 for data y = 3.0, 6.4, 9.0, 12.0. Find the sum of squared residuals SSres. [2 marks]

Related Math AI HL Topics

Continue learning with these related topics from the same unit:

4.1.1Population and Samples
4.1.2Data Classification
4.1.3Sampling Techniques
4.1.4Data Reliability and Outliers
View all Math AI HL topics

Improve your exam technique

Command terms, paper structure, and mark-scheme tips for Math AI HL

Previous
4.12.1Sampling, reliability & validity
Next
E(X), Var(X) & linear transformations4.14.1

11 practice questions on Non-linear regression

Students who practiced this topic on Aimnova scored 82% on average. Try free practice questions and get instant AI feedback.

Try 3 Free QuestionsView All Math AI HL Topics