CDT502 Principles of Data Analytics and Decision Making
The Complete Exam Bible · Term 1, 2026CDT502 Overview
- Lingnan University
- Term 1, 2026
- Postgraduate
- Statistics
- Quantitative decision methods
What this course is asking you to do
CDT502 combines core statistics with the practical discipline of decision support.
- Assessment split In-class discussion carries twenty per cent, the group presentation and report carry forty per cent, and the final exam carries forty per cent.
- Attendance rule Across course classes, the minimum attendance requirement is at least 80%; missing at least three sessions triggers a formal warning, regular attendance is mandatory for grading eligibility, and students should check the course LMS record.
- Core method Define the decision and evidence scope before choosing a statistic, probability rule, sampling design, visual or predictive model.
- Project standard Align the question, dataset, validation, interpretation and operational implication, then regenerate every reported result from the delivered files.
How CDT502 is assessed
| Component | Weight | Format |
|---|---|---|
| In-class Discussion | 20% | Attendance and active participation, live quizzes, Python exercises and classroom checks |
| Group Presentation and Report | 40% | Team analytics assignment with reproducible analysis, presentation and written report |
| Final Exam | 40% | Core concepts, analytical reasoning, method selection and interpretation of results |
The three numeric components total 100%. The published detail divides the first component into Attendance at 10% and In-class Discussion at 10%. Across course classes, the minimum attendance requirement is at least 80% attendance; missing at least three sessions triggers a formal warning, and regular attendance is mandatory for grading eligibility. Check the course LMS attendance record and current instructions. The captured source does not publish the final exam duration, sections, question types or permitted materials.
Assessment structure
The segment widths reproduce the published numeric weights and total 100%.
What CDT502 covers
Six chapters develop the captured teaching body from descriptive evidence and probability through sampling, visual communication and reproducible decision studies.
From Data to Decisions
evidence roles, descriptive and inferential claims, uncertainty and bounded recommendationsThis chapter separates observed records, descriptive summaries, population inference and operational action.
- Data, information and decision evidence
- Descriptive versus inferential claims
- Units, denominators and scope
- Mean and median under skew
- Uncertainty before action
- Bounded operational recommendations
Describing Distributions
frequency, centre, spread, shape, outliers and scale-aware summariesThis chapter reads a distribution through location, spread, shape and unusual observations.
- Frequency and relative frequency
- Mean, median and mode
- Range and interquartile range
- Variance and standard deviation
- Skew, bins and outliers
- Measurement scales and standardisation
Probability and Conditional Evidence
event rules, independence, conditional probability, trees, Bayes and base ratesThis chapter treats probability as a language for events and changing information.
- Objective and subjective probability
- Complements, unions and intersections
- Mutual exclusivity and independence
- Conditional probability
- Probability trees
- Bayes theorem and base rates
Sampling and Credible Inference
populations, sampling designs, bias, sampling distributions and confidenceThis chapter follows the route from population and parameter through sampling rule, sample and statistic.
- Population, parameter, sample and statistic
- Probability sampling designs
- Coverage and nonresponse
- Sampling error versus bias
- Sampling distributions
- Standard error and confidence
Visual Evidence and Data Stories
visual channels, chart choice, distributions, relationships, residuals and narrativeThis chapter selects charts from the analytical comparison rather than a software gallery.
- Visual attention and channels
- Chart choice by variable role
- Distribution displays
- Relationships and residuals
- Colour, facets and scales
- Narrative and publication checks
Reproducible Decision Studies
data audits, time-aware validation, baselines, error analysis and decision scopeThis chapter turns the group project's marking points into an analytical workflow. A bounded decision question drives data auditing, feature construction, time-aware validation, baseline selection, common-fold model comparison and segmented error analysis.
- Decision questions and scope
- Data audits and preprocessing
- Time-aware validation
- Transparent baselines
- Error metrics and segmentation
- Reproducibility and decision translation
The published course description moves from regression, optimisation and simulation toward one repeated outcome: generating insights from complex real-world datasets and using them to support business decisions. The captured teaching body begins earlier in that chain.
It establishes how observed records become descriptive summaries, how probability qualifies uncertain evidence, how sampling design controls population claims, how visual channels shape comparison, and how a reproducible study connects method to action.
How the chapters fit together
From Data to Decisions separates records, summaries, inference and action so that each claim has a visible warrant.
Describing Distributions then builds frequency, centre, spread, shape and outlier checks without treating one average as a complete description. Probability and Conditional Evidence makes denominators explicit through event rules, trees and Bayes reasoning. Sampling and Credible Inference distinguishes random sampling variation from coverage, nonresponse and measurement bias.
Visual Evidence and Data Stories treats a chart as an evidence mapping rather than decoration. Reproducible Decision Studies joins the chain through data auditing, time-aware validation, common-fold model comparison, error analysis and bounded operational conclusions.
What the assessment rewards
The published assessment comprises in-class discussion, a group presentation and report, and a final exam.
The group brief is particularly explicit about quality: a focused research question, a suitable credible dataset, rigorous and reproducible methods, correct interpretation, useful decision implications, coherent academic presentation and meaningful participation. Those criteria explain why this guide shows every denominator, preserves validation order and distinguishes predictive association from causal effect.
The final exam description names core concepts, analytical reasoning, method selection and interpretation, so the closing revision routine concentrates on those abilities rather than guessing an unpublished paper format.
Attendance and grading eligibility
The published attendance scope is the course classes.
The minimum attendance requirement is at least 80%; missing at least three sessions triggers a formal warning, and regular attendance is mandatory for grading eligibility. Attendance also contributes within the twenty per cent in-class component.
Check the course LMS attendance record and current instructions rather than treating the weighted mark and the eligibility condition as interchangeable.
How to study efficiently
For every worked example, predict the direction before calculating, rebuild the result from the printed numbers and finish with two sentences: what the result supports and what it does not establish.
Then change one condition and determine which parts of the conclusion survive. This makes formulas usable under new wording and exposes the recurrent errors that cost marks: reversed conditions, hidden denominators, unsupported population claims, visual comparisons with misleading scales and model rankings detached from operational loss.
Worked example · free
Select a model using the costly error
- 3Calculate each overall improvement against the baseline.
- 2Compare the two candidates during the costly Friday dinner period.
- 3Recommend one model with a scope limit and next measurement.
Key terms
- Population
- The complete set of units about which an analysis seeks to make a claim.
- Parameter
- A numerical feature of a population, such as its mean or proportion.
- Statistic
- A numerical feature calculated from observed sample data.
- Median
- The middle ordered value, or the average of the two middle values when the count is even.
- Standard Deviation
- A measure of spread in the original unit based on squared deviations from the mean.
- Conditional Probability
- The probability of an event within the restricted set where another event has occurred.
- Base Rate
- The prevalence of a condition before new evidence is observed.
- Sampling Frame
- The operational list or mechanism from which population units can be selected.
- Sampling Distribution
- The distribution of a statistic across repeated samples drawn by the same rule.
- Standard Error
- The typical sampling variation of an estimator under its design and assumptions.
- Residual
- The observed outcome minus the value fitted or predicted by a model.
- Data Leakage
- Use of information unavailable at prediction time when training, transforming or selecting a model.
- Baseline Model
- A simple reference rule that a candidate model must beat under the same evaluation protocol.
- Reproducibility
- The ability to regenerate reported results and figures from documented data, code and settings.
CDT502 FAQ
How is the course assessed?
The published plan assigns twenty per cent to in-class discussion, forty per cent to the group presentation and report, and forty per cent to the final exam. The discussion component includes attendance, live quizzes, Python exercises and classroom checks. The group brief independently confirms its forty per cent course weight.
Does attendance affect grading eligibility?
Yes. Across course classes, the minimum attendance requirement is at least 80%. Missing more than twenty per cent, defined as at least three sessions, triggers a formal warning, and regular attendance is mandatory for grading eligibility. Check the course LMS attendance record and current instructions if an absence is not reflected correctly.
What is the final exam format?
The captured material identifies a Final Exam worth forty per cent and says it assesses core concepts, analytical reasoning, method selection and interpretation of results. It does not publish a duration, section structure, question types or permitted materials. Use the current LMS announcement for those operational details rather than assuming a format.
When should I use the mean instead of the median?
Use the mean when the balance point is substantively meaningful and the distribution is not dominated by skew or influential extremes, normally with a standard deviation. Use the median with an interquartile range when resistance matters. Always inspect the distribution because neither centre describes spread, clusters or unusual cases by itself.
Why is a positive flag not the same as a high posterior probability?
A flag can occur among both target and non-target cases. The posterior probability among flags depends on the condition's base rate, sensitivity and false-positive rate. Build a tree or imagined cohort, count every route to a flag and divide true target flags by all flags. Do not reverse the conditioning direction.
How do I justify a sampling design?
Begin with the decision population and identify the frame, unit, inclusion rule, nonresponse path and measurement process. Choose stratification when subgroup coverage matters, clustering when collection cost dominates, or a transparent random rule when a usable frame exists. State which units remain uncovered and separate design bias from random sampling error.
What makes a chart analytically useful?
A useful chart maps the decision-relevant comparison to an accurate visual channel. Position on a common scale is usually strongest for close comparisons; colour is better for grouping or emphasis. State units, denominators, period, missing-value treatment and uncertainty, and avoid causal titles when the underlying design supports only association.
What does reproducible mean for the group project?
Another analyst should be able to regenerate every reported table, metric and figure from the delivered data and code. Document exclusions, feature construction, random seeds, fold boundaries, transformations and package requirements. Keep the final holdout untouched until model selection is complete and compare every candidate with the same outcome and validation folds.
Questions students ask in each CDT502 chapter
- Why is a correct average not enough for a decision?
- An average compresses observed values but does not show spread, selection, unusual cases or the cost of acting incorrectly. A decision also needs the evidence scope, uncertainty and objective. Report the average as one input, then state which action it supports and which additional evidence would change that action. Chapter 01 · From Data to Decisions
- Why can two groups with the same mean need different decisions?
- The mean describes a balance point, not variability. One group may be tightly concentrated around the mean while another alternates between low and high values. Staffing, inventory and risk decisions respond to that spread, so compare standard deviation, interquartile range and the full distribution before applying one rule. Chapter 02 · Describing Distributions
- What is the difference between independent and mutually exclusive?
- Mutually exclusive events cannot occur together, so their intersection is zero. Independent events may occur together, but observing one does not change the probability of the other. Nontrivial mutually exclusive events are therefore not independent because observing one makes the other impossible. Chapter 03 · Probability and Conditional Evidence
- Why does a large sample not guarantee representativeness?
- Sample size controls random precision after a selection process is established. If parts of the target population have zero inclusion probability or response depends on the outcome, more observations from the same mechanism repeat the distortion more precisely. Audit the frame and response path before celebrating a narrow interval. Chapter 04 · Sampling and Credible Inference
- How do I choose between a histogram and a boxplot?
- Use a histogram when distribution shape, peaks or gaps matter and record the bin rule. Use a boxplot for compact comparisons of median, middle spread and unusual values across groups. When the dataset is small or individual values matter, add a dot plot rather than asking either summary to show everything. Chapter 05 · Visual Evidence and Data Stories
- Why is a random split risky for time-indexed data?
- It can place later observations in training and earlier ones in testing, allowing future seasonal or policy information to influence a model evaluated as though the future were unknown. Use expanding or rolling windows and preserve a final later block so evaluation matches deployment direction. Chapter 06 · Reproducible Decision Studies
How to study for the exam
Study by error type rather than by slide order. First rebuild one descriptive example and check how centre, spread and shape change when an extreme value moves. Next draw probability trees until every branch and denominator can be read aloud. Then audit several sampling claims by writing the target population, frame, selection rule and uncovered units.
For visualisation, state the comparison before naming a chart and explain why its channel makes that comparison accurate. Finish with one small predictive study: preserve time order, fit transformations inside training folds, compare a baseline and candidate on identical periods, and inspect errors in the operational segment that matters most.
After every exercise, write one sentence on what the result supports and one on what it does not establish. Keep a running list of reversed conditions, denominator mistakes, causal overclaims and leakage; revisiting those mistakes is more useful than rereading examples that already feel familiar.
Confirm final exam logistics and section-specific project dates on the current LMS page because the captured materials do not publish a complete exam format and project calendars differ by section.
Your AI Statistics tutor for CDT502
Stuck on a hard CDT502 question? Sia is AskSia’s AI Statistics tutor — ask any CDT502 Principles of Data Analytics and Decision Making 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.
