BUSS6002 Data Science in Business
BUSS6002 Overview
- University of Sydney
- S2 2026
- 11 chapters
- Statistics
BUSS6002 Data Science in Business is a University of Sydney postgraduate unit that deliberately weaves together business analytics, marketing and information systems into one data-science spine, taking you from the business framing of data through the statistical and linear-algebra machinery behind clustering, regression, classification and model selection — with hands-on Python.
- Assessed by Mid-semester exam 25% · Individual assignment 30% · Final exam 45% · Pass requirement Hurdle
- Key terms Big data (4 V's), CRISP-DM, Exploratory Data Analysis (EDA), Inner product and Euclidean norm
- How to prepare Treat BUSS6002 as two courses sharing one exam.
- Most asked What mark do I need to continue to QBUS6810 or QBUS6840?
What you get in the full BUSS6002 Exam Bible
- 11 chapters, 120 pages — every examinable topic, week by week
- 20 original practice questions in the Final exam format
- 11 worked examples, solved the way BUSS6002 marks
- 80 computed figures, redrawn from the course’s own numbers
- Exam-morning plan for the Final exam (45%) — what to revise last, where marks are lost
Free on this page: assessment table, chapter map, one worked example, key terms, FAQ. The rest is in the full exam bible.
How BUSS6002 is assessed
| Component | Weight | Format |
|---|---|---|
| Mid-semester exam | 25% | In-person pen-and-paper, 1 hour, Weeks 1–6; MCQ 12 + Short answer 15 + Python code 8; 1 A4 single-sided handwritten note allowed |
| Individual assignment | 30% | Individual Python/sklearn coding task + written report; GenAI reflection required if AI used |
| Final exam | 45% | In-person pen-and-paper, 2h + 10min reading, whole semester; MCQ 12 + Short answer 27 + Python code 6 = 45 marks; 1 A4 double-sided handwritten note allowed |
| Pass requirement | Hurdle | Must obtain at least 50% overall; no component hurdle ('There are no other requirements') |
What other students are asking about BUSS6002
- Where can I find lecture notes or study guides for BUSS6002 at USyd?
- The BUSS6002 study guide in AskSia Library covers Data Science in Business in 11 chapters, with the assessment table, a chapter-by-chapter topic map, key terms and a worked example, free to read on this page.
- Is there a final exam in BUSS6002, and how much is it worth?
- Yes. In the offering this AskSia Library guide is built from, Final exam carries 45% of the BUSS6002 mark; the full split is Mid-semester exam 25%, Individual assignment 30%, Final exam 45%, Pass requirement Hurdle.
- What does BUSS6002 cover on organisational data & analytical capabilities?
- In AskSia Library's BUSS6002 guide, the Organisational Data & Analytical Capabilities chapter explains data lifecycle · learning loops · DS teams · decision environments · analytical-capability types · data types & dataset structure.
- What does BUSS6002 cover on knowledge discovery & exploratory data analysis?
- In AskSia Library's BUSS6002 guide, the Knowledge Discovery & Exploratory Data Analysis chapter explains kDDA · CRISP-DM · Snail Shell process · EDA techniques · data quality — missing data & outliers.
- What does BUSS6002 cover on linear algebra i: vectors & scientific computing?
- In AskSia Library's BUSS6002 guide, the Linear Algebra I: Vectors & Scientific Computing chapter explains vectors · vector operations · geometric intuition · NumPy scientific computing.
What BUSS6002 covers
BUSS6002 runs as eleven examinable chapters that build from the business framing of data through the linear-algebra and statistical machinery that the closed-book exams actually test — here is the full map.
Foundations of Data Science in Business
History of data in marketing · big vs small data · data as a decision tool · data-science perspectivesWeek 1 of BUSS6002 sets the conceptual spine the rest of the unit hangs on.
- Data is collected, not given — datasets reflect the choices of whoever gathered them, which seeds later bias/quality/ethics chapters
- Data science already reshapes business, most visibly in marketing (customer internet behaviour now drives strategy)
- Value comes from analysis and interpretation, not from data itself — 'big data comes with big responsibility'
- Big data = the four V's: Volume, Variety, Velocity, Veracity
- Small data — bounded, human-comprehensible, deliberately collected for a specific present-day decision
- The examinable axis: how data ARISES (passive/continuous vs deliberate/once-off), not its byte count
Organisational Data & Analytical Capabilities
Data lifecycle · learning loops · DS teams · decision environments · analytical-capability types · data types & dataset structureWeek 2 of BUSS6002 is the “business and information-systems” framing of data science, and it is where quant-anxious students leave easy marks on the table.
- Data lifecycle — collect → store → process → analyse → act → archive/feedback, run as a continuous cycle
- Learning loop — sense → decide → act → feedback, so plans are checked against reality and adjusted
- Data-science team continuum — roles from the business domain through analytics to the IT domain; always a team effort
- Decision contexts (Cynefin) — simple, complicated, complex, chaotic, each defined by its cause–effect relationship
- Decision framework — diagnoses which context you are in so you avoid a misaligned management style
- Four analytical capabilities — descriptive (what), diagnostic (why), predictive (what next), prescriptive (what to do)
Knowledge Discovery & Exploratory Data Analysis
KDDA · CRISP-DM · Snail Shell process · EDA techniques · data quality — missing data & outliersThis Week 3 chapter frames the whole knowledge-discovery process (KDDA) with two named process models — CRISP-DM, whose six phases you must recite in order, and the Snail Shell model, defined by its continuous-iteration spiral.
- KDDA — discovering valid, non-trivial, previously unknown, actionable patterns
- CRISP-DM — the six ordered phases from Business Understanding to Deployment
- Snail Shell model — a KDDA process model built on continuous iteration
- EDA — the summarise → hypothesise → model loop and its two axes
- Four sample moments — location, scale (sd), symmetry (skew), tails (kurtosis)
- Quantiles & the box-plot — five-number summary and 1.5×IQR fences
Linear Algebra I: Vectors & Scientific Computing
Vectors · vector operations · geometric intuition · NumPy scientific computingWeek 4 is where BUSS6002 turns from concepts into machinery: a dataset is a stack of vectors, and linear algebra lets you “vectorise” a statistic into one compact expression a computer evaluates fast.
- Vectors in ℝⁿ — ordered lists, columns by default, the zero 0 and ones 1 vectors
- Scalar multiplication & vector addition — entry-by-entry, equal sizes only
- Linear combinations & linear (in)dependence — the non-trivial-coefficients test
- Inner (dot) product ⟨x,y⟩ = xᵀy — a scalar, with its algebraic and angle properties
- Euclidean norm ‖x‖₂ = √⟨x,x⟩ and squared norm ‖x‖₂² (used by K-means & OLS)
- Euclidean distance ‖x − y‖₂ and orthogonality ⟨x,y⟩ = 0
Clustering & Customer Segmentation
Unsupervised learning · K-means algorithm · distance/similarity · customer segmentation applicationClustering is the unit's main unsupervised method: with no labels to predict, it partitions n data points into k groups that are similar within and different between.
- Supervised vs unsupervised — clustering learns structure with no labels
- Customer segmentation — four bases: demographic, geographic, psychographic, behavioural
- Surrounding metrics — market size & value, market share, customer lifetime value (CLV)
- Squared Euclidean distance — the only distance K-means uses, and why it skips the root
- Centroid — the per-coordinate mean of a cluster's members (not a fixed data point)
- Clustering objective (WCSS) — sum of squared point-to-centroid distances, minimised
Matrices & Linear Regression
Matrix operations · OLS in matrix form · interpretation · regression diagnostics · scientific computing IIWeek 6 turns the unit's linear-algebra notation into its single most-examined model.
- The data matrix — rows = observations, columns = variables; bold UPPER-case notation
- Matrix operations — transpose, identity, inverse, determinant (element-wise add/scalar)
- The matrix product — row·column, the inner-dimension rule, and why AB ≠ BA
- The regression model — simple y = β₀ + β₁x + ε and matrix form y = Xβ + ε
- OLS as the closed-form RSS minimiser — β̂ = (XᵀX)⁻¹Xᵀy
- Coefficient interpretation — an associated average change, never a cause
Feature Engineering & Data Ethics
Ethical & legal issues · responsible AI · privacy/fairness · feature transformations · unstructured/text dataWeek 7 is two examinable halves in one chapter.
- Five ethics principles — Ownership, Transparency, Privacy, Intention, Outcomes, kept distinct
- Disparate impact & PII — unintended group harm is unlawful even with good intent
- Legal landscape — HiQ v. LinkedIn (public scraping ≠ hacking) and the Robodebt disaster
- Responsible AI — 'AI reflects the values of its creators', black-box opacity, deep-fakes
- Feature engineering — re-express variables so a linear model fits a non-linear relationship
- The transform toolkit — log response, interaction, polynomial, linear basis functions
Classification & Logistic Regression
Classifiers · evaluation metrics · logistic regression · log-odds & sigmoid · predictive analytics in marketingThis is the Week 8 block where BUSS6002 stops predicting numbers and starts predicting categories — will a customer churn, is a transaction fraud, will a lead convert.
- Classifier vs probabilistic classifier — label f(x) vs a predicted probability P̂(y|x)
- The decision threshold τ — default 0.5, equal-cost assumption (BUSS6002 does not optimise it)
- The confusion matrix — TP, FN, FP, TN as a 2×2 table of prediction vs truth
- Metrics off the matrix — accuracy, recall (TPR), specificity (TNR), FPR, precision, F1
- The GLM template — g(E(y|x)) = xᵀβ, identity link = linear regression
- Logistic regression — logit link log(p/(1−p)) = xᵀβ and the sigmoid activation
Model Evaluation & Selection
Measuring marketing-campaign success · bias–variance tradeoff · validation MSE · selecting the optimal modelWeek 9 of BUSS6002 is the quantitative heart of the unit's second half and a near-guaranteed block on the cumulative final exam.
- Linear basis function (LBF) models — y = phi(x)^T beta + eps; flexibility grows with the number of basis functions p
- Polynomial vs radial basis functions — phi_i(x)=x^i versus a Gaussian bump exp(-(x-c_i)^2 / 2s^2)
- Design matrix and least squares — fit beta_hat = (Phi^T Phi)^-1 Phi^T y, the same normal equations as linear regression
- Model selection — estimate each candidate's performance on unseen data and pick the best complexity
- Expected Prediction Error — EPE = E[(y - yhat)^2] under squared loss, a measure of generalisation
- Bias-variance decomposition — EPE = irreducible sigma^2 + Bias^2 + Variance, only the last two reducible
Maximum Likelihood & Optimisation
Maximum Likelihood Estimation · log-likelihood · analytic vs iterative solutions · gradient descentWeek 10 pulls back the curtain on how a model's parameters are actually estimated.
- Estimation is optimisation — argmin/argmax is the estimate; argmin (location) vs min (value)
- Likelihood — L(θ) = p(data|θ) = ∏ p(yᵢ|θ) for an independent sample
- Maximum likelihood — θ̂ = argmax L(θ): the value that makes the data most plausible
- Log-likelihood — ℓ(θ) = Σ log p(yᵢ|θ); why we log (underflow, product→sum, monotonic)
- The MLE recipe — log → differentiate → set = 0 → solve → check 2nd derivative < 0
- Worked distributions — Bernoulli L = θ^h(1−θ)^t; Poisson MLE = sample mean; exponential
Big Data Solutions
Wide vs tall data · algorithm complexity · scalable computational approachesWeek 11 reframes big data as a computation problem rather than a storage one: when n or p reaches the millions, the same estimators (OLS, logistic regression) become too slow or impossible to fit directly.
- Big data as a compute problem — the 4 V's (Volume, Variety, Velocity, Veracity) resurfacing as a bottleneck
- Wide data — large p (many predictors, few observations); fix = reduce predictors / less-flexible model
- Tall data — large n (many observations, few predictors); fix = row-blocks / stochastic gradient descent
- Why wide breaks OLS — p > n overfits and XᵀX becomes singular (no unique inverse)
- Big-O notation — an asymptotic upper bound on worst-case running time vs input size n
- The growth ladder — O(1) < O(log n) < O(n) < O(n log n) < O(n²) < O(2ⁿ)
Across eleven examinable chapters you cover big-data thinking and the CRISP-DM knowledge-discovery process, EDA, vectors and matrices, K-means segmentation, linear and logistic regression, feature engineering, data ethics, the bias–variance tradeoff, maximum likelihood and gradient descent, and big-data scalability.
The catch is that it looks like a business unit but is examined like a quantitative-methods unit: a 25% mid-semester exam, a 30% coded assignment, and a 45% closed-restricted final with hand-written Python and short-answer derivations.
The exam rewards three things — doing the maths by hand in the course's exact notation, diagnosing residual plots and selecting models via bias–variance, and writing runnable NumPy/sklearn from memory — while still paying out the easy framework marks that under-prepared students drop.
A result of 65 (not just the 50% pass) is the real bar for anyone continuing to QBUS6810/6840.
BUSS6002 cheatsheet for the final exam
AskSia publishes a free BUSS6002 cheatsheet for the final exam, a handwritten A4 planner that follows the official permitted-materials rule for this assessment and condenses the topics in this guide.
Worked example · free
Logistic regression: predicted probability and the decision threshold
- +1State the model and compute the linear predictor (the log-odds): η = β₀ + β₁x₁ + β₂x₂ = −2.0 + (2.0)(1.5) + (1.0)(1) = −2.0 + 3.0 + 1.0 = 2.0. This score lives on the log-odds scale, not the probability scale.
- +1Map the log-odds to a probability with the sigmoid (the inverse-logit activation): P̂(y = 1 | x) = 1 / (1 + exp(−η)) = 1 / (1 + exp(−2.0)).
- +1Evaluate numerically: exp(−2.0) ≈ 0.135, so P̂(y = 1 | x) = 1 / (1 + 0.135) = 1 / 1.135 ≈ 0.88.
- +1Apply the default threshold τ = 0.5: because 0.88 > 0.5, classify the customer as y = 1 (predicted to churn). Equivalently, a positive log-odds (η > 0) always gives P̂ > 0.5.
Key terms
- Big data (4 V's)
- Data characterised by Volume, Variety, Velocity and Veracity, in contrast to deliberately collected small data; the unit stresses that value comes from analysis, not from size alone.
- CRISP-DM
- The canonical knowledge-discovery process model with six phases: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation and Deployment. Knowing the order and purpose of each phase is a recurring exam ask.
- Exploratory Data Analysis (EDA)
- Iteratively summarising and visualising data (sample moments, quantiles, histograms, box-plots, Q-Q plots) to find patterns, spot anomalies and test assumptions before modelling.
- Inner product and Euclidean norm
- The dot product ⟨x, y⟩ = Σ xᵢyᵢ = xᵀy returns a scalar; the 2-norm ‖x‖₂ = √⟨x, x⟩ measures length. Distance between points is ‖x − y‖₂ — the building block for K-means and OLS.
- K-means clustering
- An unsupervised algorithm that partitions points into k groups by alternating an assignment step (nearest centroid) and an update step (recompute centroid means), minimising within-cluster sum of squares. It converges to a local, not global, optimum.
- OLS estimator
- The ordinary-least-squares coefficient vector in matrix form, β̂ = (XᵀX)⁻¹Xᵀy — the closed-form minimiser of the residual sum of squares for linear regression.
- Logistic regression (sigmoid and logit)
- A GLM for a Bernoulli response: the logit link sets log(p/(1−p)) = xᵀβ, and the sigmoid activation p = 1/(1 + exp(−xᵀβ)) maps the log-odds back to a probability. Fitted by maximum likelihood, not least squares.
- Confusion matrix and metrics
- The 2×2 table of TP, FP, FN, TN from which accuracy, precision, recall (TPR), specificity (TNR), FPR and F1 are computed. Accuracy is misleading under class imbalance.
- Bias–variance tradeoff
- Expected prediction error decomposes as irreducible σ² + Bias² + Variance. Flexible models cut bias but raise variance (overfit); simple models do the reverse (underfit). Model selection balances the two.
- Validation MSE
- The mean squared error on a held-out validation set, used as the practical estimate of expected prediction error to choose model complexity; the test set gives the final unbiased estimate. Training MSE is never a selection criterion.
- Maximum likelihood and gradient descent
- MLE chooses parameters that maximise the (log-)likelihood of the observed data. When there is no closed form (e.g. logistic regression), gradient descent iterates θ_{k+1} = θ_k − α∇f(θ_k) to minimise the negative log-likelihood.
- Wide vs tall data and Big-O
- Wide data has large p (many predictors) → reduce dimensions or simplify the model; tall data has large n (many observations) → block it or use stochastic gradient descent. Algorithmic cost is reasoned about with Big-O notation.
BUSS6002 FAQ
How is BUSS6002 assessed?
Three components: a 25% in-person mid-semester exam (1 hour, covering Weeks 1–6), a 30% individual Python/sklearn assignment with a written report, and a 45% in-person final exam covering the whole semester. The pass requirement is 50% overall with no individual-component hurdle.
Is there a final exam?
Yes — a 45% in-person, pen-and-paper final (2 hours writing plus 10 minutes reading) over all weeks. It mixes 12 marks of multiple choice, 27 marks of short answer with full working, and 6 marks of hand-written Python code. You may bring one A4 double-sided handwritten note sheet, a non-programmable calculator and a dictionary.
What part of the course is hardest?
The trap is that BUSS6002 looks like a business unit but is examined like a quantitative-methods unit. Students over-focus on the linear algebra, OLS, MLE and gradient descent and then throw away the easy framework marks (CRISP-DM, decision environments, data ethics). The marquee tricky results are residual-plot diagnosis, the bias–variance tradeoff, and the log-linear retransformation bias.
How should I prepare for the exams?
Practise doing the maths by hand in the course's exact notation, since notation is graded. Drill the recurring question types — vector/matrix arithmetic, K-means assignment, OLS by hand, confusion-matrix metrics, logistic probabilities, MLE derivations and Big-O reasoning — and rehearse writing runnable NumPy/sklearn from memory. Build your A4 note sheet around the formulas you reach for most.
How much maths and coding does it really involve?
A lot of both. Expect vector and matrix algebra, OLS in matrix form, Euclidean distance, logistic log-odds and sigmoid, the bias–variance decomposition, maximum likelihood and gradient-descent update rules — plus hand-written Python (NumPy/sklearn) on the exam and a coded assignment.
What mark do I need to continue to QBUS6810 or QBUS6840?
BUSS6002 is a feeder unit: a result of 65 is required to progress to QBUS6810 and QBUS6840, so treat 65 — not just the 50% pass mark — as the real bar if you plan to continue in business analytics.
Is this guide official or affiliated with the University of Sydney?
No. This is an independent AskSia study resource for BUSS6002. It is not produced, endorsed by or affiliated with the University of Sydney; always confirm assessment details, dates and policies against your official Canvas unit outline.
Can I use generative AI in the assignment?
The assignment permits AI use but requires a generative-AI reflection if you used it. Follow the unit's declaration rules exactly, and remember the exams are closed-restricted and hand-written, so your underlying skills still have to be exam-ready.
Questions students ask in each BUSS6002 chapter
- Is Week 1 actually examinable, or just background?
- It is directly examinable. Foundations shows up as 1-mark MCQ classification items (big vs small data) and short, definition-style short-answer questions in both the 25% mid-semester and the cumulative 45% final. They are among the cheapest marks in the unit, which is exactly why under-prepared students drop them. Chapter 01 · Foundations of Data Science in Business
- Is Week 2 actually examinable or just background?
- It is directly examinable and is among the cheapest marks in BUSS6002. It shows up as 1-mark MCQ classification items (analytics type, decision context, data type) and short-answer definition or justify questions in both the 25% mid-semester and the cumulative 45% final. Students who over-focus on linear algebra routinely drop these easy marks. Chapter 02 · Organisational Data & Analytical Capabilities
- Do I have to memorise the CRISP-DM phases in order?
- Yes. The exam asks you to sequence the six phases or to place a task in the right one. Learn them in order — Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, Deployment — and one job each. Exploring data and checking metadata is Data Understanding; cleaning and building derived features is Data Preparation. Chapter 03 · Knowledge Discovery & Exploratory Data Analysis
- Is the inner product a vector or a number?
- A single number (a scalar). The dot product multiplies two equal-length vectors entry-by-entry and sums the results, collapsing both n-vectors into one value. If your answer is still a vector you have done element-wise (Hadamard) multiplication instead — a different operation. This is one of the most common one-mark slips in the unit. Chapter 04 · Linear Algebra I: Vectors & Scientific Computing
- Is clustering supervised or unsupervised?
- Unsupervised — there is no label y to predict. K-means learns the grouping purely from the feature values, judged by how compact the clusters are, not by matching a known answer. Chapter 05 · Clustering & Customer Segmentation
- Is matrix multiplication commutative?
- No. In general AB ≠ BA — order matters, and assuming otherwise is a guaranteed wrong MCQ answer. The product is also only defined when the inner dimensions match (columns of A = rows of B); otherwise it is undefined. And it is never element-wise: each entry is a row of A dotted with a column of B. Chapter 06 · Matrices & Linear Regression
- How do I keep the five ethics principles straight in the exam?
- Run them as a checklist in order — Ownership (was there consent to collect?), Transparency (were subjects told how data is used?), Privacy (was PII exposed?), Intention (why is it being collected?), Outcomes (did it cause group harm?). Eliminating the satisfied principles isolates the breach. Watch for cases where consent was given but data was published — that is Privacy, not Ownership. Chapter 07 · Feature Engineering & Data Ethics
- Is the linear predictor xᵀβ a probability?
- No — this is the most punished mistake in the chapter. xᵀβ is on the log-odds scale. To get a probability you must push it through the sigmoid: P̂ = 1/(1+exp(−xᵀβ)). Sign check: a positive log-odds always gives P̂ > 0.5, and the more positive it is, the closer the probability gets to 1. Chapter 08 · Classification & Logistic Regression
- Is Model Evaluation & Selection on the final exam?
- Yes. It is Week 9 content, so it is examined in the cumulative 45% final (which covers all weeks); it is not on the 25% mid-semester, which stops at Week 6. The released sample paper shows a model-selection true/false MCQ and a bias-variance short-answer as recurring items, and a Python item can ask you to compute a validation MSE from arrays. Chapter 09 · Model Evaluation & Selection
- Why do we maximise the log-likelihood instead of the likelihood?
- Three reasons, all examinable. (1) Numerical underflow: a product of thousands of probabilities (each < 1) collapses to ~0 in floating point, while summing their logs does not. (2) The log turns the product ∏ᵢ p(yᵢ|θ) into a sum Σᵢ log p(yᵢ|θ), which is far easier to differentiate. (3) The log is strictly increasing (monotonic), so it does not move the optimum — argmax L = argmax ℓ. As a bonus, for exponential-family densities the log cancels the exponential. Chapter 10 · Maximum Likelihood & Optimisation
- What is the difference between wide and tall data?
- Wide data has a large number of predictors p but few observations n (p > n) — a short, fat matrix, like genomics with tens of thousands of gene columns for a handful of patients. Tall data has few predictors but a huge number of observations n (n ≫ p) — a long, narrow matrix, like every transaction a platform logs. The two need opposite fixes, so identifying which dimension is big is the whole task. Chapter 11 · Big Data Solutions
How to study for the exam
Treat BUSS6002 as two courses sharing one exam.
First, lock the framework chapters (Foundations, Organisational Data, Knowledge Discovery/EDA, and Feature Engineering & Data Ethics) early — they are the cheap, high-yield MCQ and short-answer marks that quant-anxious students leave on the table; memorise the 4 V's, the CRISP-DM phases, the four analytics types, the Cynefin contexts and the five ethics principles as crisp checklists.
Second, build genuine fluency in the quantitative spine by hand: vector and matrix operations, the OLS estimator (XᵀX)⁻¹Xᵀy, K-means, the logistic log-odds/sigmoid, the bias–variance decomposition, validation MSE, MLE derivations and gradient descent, all written in the course's exact notation because notation is graded.
Drill past-style questions until the recurring patterns (residual-plot diagnosis, retransformation bias, model-selection true/false, Big-O ranking) are automatic, rehearse writing runnable NumPy/sklearn from memory, and curate your single A4 note sheet around the formulas and one-line definitions you reach for most. Aim for 65, not 50, if you intend to progress to QBUS6810/6840.
Your AI Statistics tutor for BUSS6002
Stuck on a hard BUSS6002 question? Sia is AskSia’s AI Statistics tutor — ask any BUSS6002 Data Science in Business question and get a clear, step-by-step explanation grounded in how the course is actually taught and assessed. Read this whole study guide free, then take your hardest questions to Sia.