STAT5003 Computational Statistical Methods
The Complete Exam Bible · S2 2026STAT5003 Overview
- The University of Sydney
- Semester 2, 2026
Within The University of Sydney, STAT5003 Computational Statistical Methods covers computational statistics carried out in R across the whole supervised/unsupervised map: regression and local smoothing (moving average and kNN smoothers, regression splines), maximum likelihood and kernel density estimation, high-dimensional visualisation and clustering (PCA, t-SNE, MDS, k-means, hierarchical), classification (logistic regression, LDA, kNN, SVM), the bias–variance decomposition with k-fold, repeated and nested cross-validation, feature and model selection (best-subset and stepwise, Cp/AIC/BIC, ridge and lasso), trees and ensembles (bagging, random forests, AdaBoost/stochastic gradient boosting/XGBoost), bootstrap resampling and missing-data mechanisms and imputation (MCAR/MAR/MNAR, single, multiple, MICE), and finally Monte Carlo integration (inverse transform,
acceptance–rejection) and Markov chain Monte Carlo (Metropolis–Hastings, Gibbs).
The exam does not ask you to write R — it asks you to read it.
- STAT5003 grading 60% invigilated final exam, 35% group project (Week 4 sign-up, Week 7 Plan & EDA, Week 12 presentation + final report, Week 13 peer contribution review), 5% weekly workshop contribution (0.5 mark per workshop, capped at 5)
- STAT5003 task mode Invigilated in-person final exam — Thursday 11 June 2026, 9:00am, venue per personal exam timetable. Structure: 12 multiple-choice questions worth 24 marks (each has two correct answers; 1 mark per correct selection, no deduction for incorrect or unanswered) + 4 equally weighted extended-answer questions worth 36 marks, written in the answer book provided. Permitted: one A4 double-sided HANDWRITTEN cheat sheet (printed or iPad-printed sheets may be confiscated or referred for academic integrity review; notes may be in your native language, colour highlighters allowed), a non-programmable calculator and a bilingual dictionary.
- STAT5003 mark trap Two places. (1) The MCQ format: every question has exactly two correct answers and marks are awarded only for correct selections, so a student who picks one safe option per question caps out at 12 of 24 marks. (2) The group project's dataset gate. Reports are also mark-docked for hiding work: Use code folding so markers can view the code when needed.
- STAT5003 pass rule There is no double-pass requirement for this unit - an overall final mark of 50 is sufficient to pass the unit. Workshop marks are attendance-linked: You will be given 0.5 mark per workshop (up to a maximum of 5 marks), if you actively contribute to all workshop activities. If you miss a workshop, you can apply for special consideration - if eligible, you will be given a mark adjustment for the final exam.
How STAT5003 is assessed
| Component | Weight | Format |
|---|---|---|
| Weekly Workshop Contribution | 5% | Participation monitored in allocated workshops |
| Group Project | 35% | Collaborative analysis of a selected dataset |
| Invigilated Final Examination | 60% | Individual secure examination |
The three current components total 100%. Retrieved exam guidance says students interpret R output rather than write R code, may use a handwritten double-sided A4 cheat sheet, a bilingual dictionary and non-programmable calculator, and are not given a formula sheet; verify the current S2 exam notice.
What STAT5003 covers
The learning path moves from Assessment Map and Reproducible Workflow, through the problems opened by Cross-Validation and Model Selection, to the synthesis required in Project and Final-Exam Synthesis.
Assessment Map and Reproducible Workflow
5/35/60 structure · R workflow · assumption and output audit · connect every statistical claim to code, output, diagnostic and interpretation02Data Objects, Visualisation and Simulation
data structures · graphics · random simulation · use computation to inspect structure before fitting a model03Regression Computation and Interpretation
linear model · prediction · residual diagnostics · translate coefficient output into a conditional prediction and uncertainty statement04Density Estimation and Distribution Shape
histogram and kernel density · bandwidth · distribution comparison · evaluate how smoothing choices change the visible structure05Classification and Nearest Neighbours
classification rule · distance and scaling · confusion matrix · connect a classification threshold to errors and stakeholder cost06Missing Data and Support Vector Machines
missingness mechanism · imputation boundary · margin and kernel · separate data-loss assumptions from the classifier fitted after preprocessing07Cross-Validation and Model Selection
training and validation · k-fold cross-validation · tuning bias · estimate out-of-sample performance without leaking validation information08Trees, Ensembles and Variable Importance
decision tree · random forest · importance measure · compare predictive gain with stability and interpretability09Bootstrap and Resampling Inference
empirical resampling · bootstrap distribution · interval construction · approximate sampling uncertainty from a reproducible resampling scheme10Monte Carlo Integration and Variance
Monte Carlo estimator · simulation error · variance reduction · quantify approximation error and improve efficiency without changing the target11Bayesian Computation and MCMC
prior and likelihood · posterior · Markov chain diagnostics · separate posterior updating from the computation used to approximate it12Project and Final-Exam Synthesis
data-analysis narrative · output interpretation · assumption-sensitive conclusion · turn code and output into a concise defensible result under project or exam constraints60% of the mark is one invigilated paper of 12 multiple-choice questions that each carry exactly two correct answers plus 4 equally weighted extended-answer questions, with one handwritten double-sided A4 sheet allowed and no formula sheet supplied.
The other 35% is a semester-long group classification project run through a named Data Investigation Process with hard gates (a dataset you source yourself that must demonstrably meet at least 3 of Large/Messy/Complex/Integrated/Multi-class, at least 5 candidate classification models, a 6-page HTML plan+EDA, a 10-page final report, a 20-slide 7-minute presentation), and just 5% for contributing in workshops.
NOTE for the builder: faculty/school, level, credit points, core/elective status and prerequisites are NOT stated anywhere in the verified course material — do not assert them.
Assessment in STAT5003 is distributed as follows: 60% invigilated final exam, 35% group project (Week 4 sign-up, Week 7 Plan & EDA, Week 12 presentation + final report, Week 13 peer contribution review), 5% weekly workshop contribution (0.5 mark per workshop, capped at 5)
The operational assessment conditions matter here.
Invigilated in-person final exam — Thursday 11 June 2026, 9:00am, venue per personal exam timetable. Structure: 12 multiple-choice questions worth 24 marks (each has two correct answers; 1 mark per correct selection, no deduction for incorrect or unanswered) + 4 equally weighted extended-answer questions worth 36 marks, written in the answer book provided.
Permitted: one A4 double-sided HANDWRITTEN cheat sheet (printed or iPad-printed sheets may be confiscated or referred for academic integrity review; notes may be in your native language, colour highlighters allowed), a non-programmable calculator and a bilingual dictionary.
What makes STAT5003 demanding is concrete: Two places.
(1) The MCQ format: every question has exactly two correct answers and marks are awarded only for correct selections, so a student who picks one safe option per question caps out at 12 of 24 marks. (2) The group project's dataset gate.
Reports are also mark-docked for hiding work: Use code folding so markers can view the code when needed.
There is no double-pass requirement for this unit - an overall final mark of 50 is sufficient to pass the unit. Workshop marks are attendance-linked: You will be given 0.5 mark per workshop (up to a maximum of 5 marks), if you actively contribute to all workshop activities.
If you miss a workshop, you can apply for special consideration - if eligible, you will be given a mark adjustment for the final exam.
The learning path moves from Assessment Map and Reproducible Workflow, through the problems opened by Cross-Validation and Model Selection, to the synthesis required in Project and Final-Exam Synthesis.
Worked example: Computational Statistical Methods integrated response
- 1Extract the outcome, actor or operation that the Computational Statistical Methods integrated response task actually requires.
- 1State the precondition under which 5/35/60 structure is relevant rather than merely familiar.
- 1Use histogram and kernel density to reject the nearest alternative, then run a failure-path check with decision tree.
- 1Choose the response and state when it must be withdrawn or narrowed: Importance is model- and metric-dependent, not a causal ranking.
Key terms
- multiple linear regression
- Multiple linear regression models the conditional mean of a response as an intercept plus coefficients multiplying two or more predictors, with each coefficient interpreted holding the others constant under stated assumptions.
- bias–variance decomposition
- Bias–variance decomposition separates expected prediction error into irreducible noise, squared systematic bias and variance caused by sensitivity to the training sample.
- k-fold, repeated and nested cross-validation (nested CV prevents data leakage)
- K-fold cross-validation rotates validation across data folds, repetition reduces split sensitivity, and nested cross-validation separates inner model tuning from outer performance estimation to prevent leakage.
- ridge and lasso regularisation and the tuning parameter λ
- Ridge adds an L2 squared-coefficient penalty and lasso an L1 absolute-coefficient penalty to the loss; λ controls shrinkage, with lasso capable of setting coefficients exactly to zero.
- best-subset and stepwise selection; Cp, AIC, BIC, adjusted R²
- Best-subset and stepwise procedures search predictor sets, while Cp, AIC, BIC and adjusted R² balance goodness of fit against model complexity using different penalties.
- kernel density estimation and bandwidth h; maximum likelihood estimation
- Kernel density estimation builds a smooth distribution estimate by centring kernels on observations, with bandwidth h controlling smoothness; maximum likelihood selects parameter values that maximise the observed-data likelihood.
- confusion-matrix metrics
- Confusion-matrix metrics derive from true positives, false positives, true negatives and false negatives, including accuracy, sensitivity or recall, specificity, precision and related trade-offs.
- support vector machines
- A support vector machine chooses a maximum-margin separating boundary determined by support vectors and can use kernels to represent nonlinear boundaries in a transformed feature space.
STAT5003 FAQ
Is STAT5003 hard?
Two places. (1) The MCQ format: every question has exactly two correct answers and marks are awarded only for correct selections, so a student who picks one safe option per question caps out at 12 of 24 marks. (2) The group project's dataset gate. Reports are also mark-docked for hiding work: Use code folding so markers can view the code when needed.
How is STAT5003 assessed?
60% invigilated final exam, 35% group project (Week 4 sign-up, Week 7 Plan & EDA, Week 12 presentation + final report, Week 13 peer contribution review), 5% weekly workshop contribution (0.5 mark per workshop, capped at 5)
What is the STAT5003 exam or final-task format?
Invigilated in-person final exam — Thursday 11 June 2026, 9:00am, venue per personal exam timetable. Structure: 12 multiple-choice questions worth 24 marks (each has two correct answers; 1 mark per correct selection, no deduction for incorrect or unanswered) + 4 equally weighted extended-answer questions worth 36 marks, written in the answer book provided.
Permitted: one A4 double-sided HANDWRITTEN cheat sheet (printed or iPad-printed sheets may be confiscated or referred for academic integrity review; notes may be in your native language, colour highlighters allowed), a non-programmable calculator and a bilingual dictionary.
Does STAT5003 have a hurdle or component-level pass rule?
There is no double-pass requirement for this unit - an overall final mark of 50 is sufficient to pass the unit. Workshop marks are attendance-linked: You will be given 0.5 mark per workshop (up to a maximum of 5 marks), if you actively contribute to all workshop activities. If you miss a workshop, you can apply for special consideration - if eligible, you will be given a mark adjustment for the final exam.
What prerequisites or restrictions apply to STAT5003?
Check the current official handbook before enrolling in STAT5003; prerequisites are not inferred from its course number.
Is STAT5003 offered in Semester 2, 2026?
This resource is aligned to Semester 2, 2026. Confirm your class and assessment timetable in the current institutional system.
Is this STAT5003 resource an official university guide?
No. It is an independent STAT5003 study resource; current institutional instructions remain authoritative for assessment operation.
How to study for the exam
Retrieve the unit map, practise the recurring method—make the data-generating assumption, computation, diagnostic and uncertainty interpretation visible before selecting a statistical conclusion—on changed scenarios, and verify every operational assessment detail in the live institutional system.
Your AI Statistics tutor for STAT5003
Stuck on a hard STAT5003 question? Sia is AskSia’s AI Statistics tutor — ask any STAT5003 Computational Statistical Methods 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.
