FIT1043 · Introduction to Data Science
Data Analysis Theory
Week 5 of Monash FIT1043 Introduction to Data Science steps back to the theory of analysis: the difference between predictive and descriptive modelling, supervised vs unsupervised learning, and the idea of a model with features (independent variables) predicting a target (dependent variable). It introduces linear regression as the first concrete predictive model — fitting a line, reading slope, intercept and the correlation r — and the caution against extrapolating beyond the linear range. Test 1 falls around this week and covers Weeks 1-4; the modelling ideas here are re-examined on the final.
What this chapter covers
- 01Machine learning: analysing data to perform a task from patterns and inference rather than explicit instructions; pattern detection is central
- 02Predictive vs descriptive modelling; the general pattern vs outlying points
- 03Supervised vs unsupervised learning; features (independent/predictor) vs target (dependent/response) variable
- 04Linear regression: fitting y = slope * x + intercept to detect a linear relationship
- 05Reading the fit: slope (direction/rate), intercept, and r (correlation) — magnitude near 1 or -1 = strong, sign = direction
- 06Python: linregress(x, y) returning slope, intercept, r_value, p_value, std_err; predicted line = [slope*xi + intercept]
- 07Prediction for a new input, and the extrapolation caveat: do not predict where the relationship stops being linear
- 08Noise: real data does not lie exactly on the true model
Fit a line, read r, predict a new value, and judge extrapolation
- +1(a) Predict with y = slope*x + intercept = 0.858*41 + 2.028 = 35.178 + 2.028 = 37.206, i.e. about 37.2 thousand. (b) r = 0.892 is close to 1 and positive, so there is a strong positive linear relationship: income rises fairly tightly with age over the fitted range.
- +1(c) No. Age 70 lies outside the fitted range (20-60), and the relationship is stated to stop being linear beyond 60 (income no longer rises). Extrapolating the straight line there would give a misleadingly high figure — a linear model is only trustworthy inside the range where the linear assumption holds.
Key terms
- Machine learning
- A method of data analysis that performs a task from patterns in existing data and inference, rather than from explicit hand-written instructions.
- Supervised vs unsupervised
- Supervised learning trains on labelled examples to predict a target (regression or classification); unsupervised learning (e.g. k-means) finds structure with no target label.
- Independent vs dependent variable
- The independent (predictor/feature) variable is the input used to predict the dependent (response/target) variable, e.g. Age (independent) predicting Income (dependent).
- Linear regression
- A linear approach to model the relationship between a response and one or more predictors, fitting y = slope * x + intercept (least squares).
- Correlation r
- The linear-correlation coefficient: magnitude near 1 (or -1) means a strong linear relationship, near 0 means weak; the sign gives the direction (positive or negative).
- Extrapolation caveat
- The rule that a fitted line is only reliable inside the range where the linear assumption holds; predicting far outside that range (or past where the trend flattens) is unsafe.
Data Analysis Theory FAQ
What is the difference between supervised and unsupervised learning?
Supervised learning uses labelled data — every training example has a known target — to learn to predict that target for new inputs; classification (categorical target) and regression (continuous target) are both supervised. Unsupervised learning has no target label and instead finds structure in the data, such as k-means grouping similar points into clusters.
How do I read the r value from a regression?
Read r two ways: its magnitude gives the strength of the linear relationship (close to 1 or -1 is strong, close to 0 is weak) and its sign gives the direction (positive means y rises with x, negative means it falls). In the unit's Age-Income lab r was about 0.89 — a strong positive relationship.
Why can't I extrapolate a regression line?
Because a linear model only captures the relationship over the range you fitted it to. If income rises with age up to about 60 and then flattens, the straight line keeps climbing and would badly over-predict a 70-year-old's income. Only predict inside the fitted range where the linear assumption still holds.
Is Week 5 examined in Test 1?
Test 1 is scheduled around Week 5 but examines the learning outcomes of Weeks 1-4, so the Week 5 modelling theory is primarily assessed in the final. Still, the regression ideas here connect straight into Week 6, so learn them properly now rather than treating them as post-test material. Confirm the exact Test 1 scope on Moodle.
Can AI help me understand regression in FIT1043?
Yes. Sia can walk through fitting a line, interpreting slope/intercept/r, predicting a new value and judging whether an extrapolation is safe, step by step, and set you fresh practice in the same style. It explains the method and checks your reasoning; it does not do graded work for you, and Monash academic-integrity rules apply. Confirm details on Moodle.
Exam move
Anchor Week 5 on one worked skill: given a slope, intercept and r, predict a new value, interpret r by magnitude and sign, and state whether the prediction is a safe interpolation or an unsafe extrapolation. Practise on the Week 5 lab numbers (slope ~0.86, intercept ~2.03, r ~0.89) and on the negative-relationship practice (r ~ -0.89) so you can read both directions. Keep the vocabulary crisp — supervised vs unsupervised, independent vs dependent, model/feature/target — because those definitions are one-mark MCQ fodder. This week bridges into Week 6's bias-variance material, so make sure the linear-model basics are automatic before you add polynomial orders.
Working through Data Analysis Theory in FIT1043? Sia is AskSia’s AI Information Technology tutor — ask any FIT1043 Data Analysis Theory question and get a clear, step-by-step explanation grounded in how FIT1043 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
