INFS3822 Artificial Intelligence for Business Analytics
The Complete Study & Assessment Guide · T3 2026INFS3822 Overview
- UNSW Sydney
- Term 3 2026
- Level 3
- Coursework assessed
INFS3822 Artificial Intelligence for Business Analytics is a Level 3 UNSW Sydney course offered in Term 3 2026. It teaches students to connect artificial intelligence and machine learning to real business decisions rather than treating an algorithm as an end in itself.
- Assessed by Analytical brief, solution proposal, team proof-of-concept, quizzes and seminar engagement
- Core method Frame the decision, prepare evidence, compare models, evaluate errors and monitor action
- AI use Follow the published rule for each component and retain a reviewable process record
How INFS3822 is assessed
| Component | Weight | Format |
|---|---|---|
| Individual machine-learning analysis | 25% | Individual analytical brief applying supervised and unsupervised machine learning to GlobalStore returns data; due Week 5 |
| Individual AI solution proposal | 20% | Individual consultancy proposal with business framing and an exploratory SAS Viya AutoML pipeline; due end of Week 7 |
| Team AI proof-of-concept project | 40% | Group report 15%, group presentation 15% and individual reflection 10%; main deliverables Week 10 and reflection Week 11 |
| Preparation and Participation | 15% | Weekly in-class quizzes 4.5% and seminar engagement 10.5%; lowest eligible week dropped |
Current INFS3822 dates
| Date | Item | Control |
|---|---|---|
| 16 October 2026 at 5:00 pm AEDT | Individual machine-learning analysis | Individual analytical brief worth 25%. |
| 30 October 2026 at 5:00 pm AEDT | Individual AI solution proposal | Individual consultancy proposal worth 20%. |
| 17 November 2026 at 5:00 pm AEDT | Team report and pitch deck | Main team deliverables inside the 40% proof-of-concept project. |
| 23 November 2026 | Individual reflection | Individual reflection inside the team project. |
Dates are as published in the Term 3 2026 assessment guide. Confirm exact deadlines and submission settings in the live LMS.
What INFS3822 covers
The course moves from artificial intelligence foundations and machine-learning task families into a six-stage business analytics lifecycle. It then develops problem framing, data preparation, regression, classification, trees, ensembles, evaluation, deep learning and unstructured-data workflows before connecting generative and agentic systems to responsible evidence and accountable business action.
AI Foundations and Business Value
Separate the business decision, evidence, learned pattern, model output and monitored action.Artificial intelligence creates business value only when a learned pattern improves a real decision. This chapter separates the organisational goal, evidence, model output, action and monitoring loop.
- Business decision before algorithm
- Evidence, model output and action
- Prediction versus judgement
- Technical, operational and ethical tests
- Monitored value and feedback
- Testable AI decision claims
Learning Paradigms and ML Tasks
Route labels, rewards and output shapes to supervised, unsupervised, semi-supervised and reinforcement learning.Learning paradigms differ by the feedback supplied during training. Labelled examples support supervised prediction, unlabelled examples support pattern discovery, mixed labels support semi-supervised learning, and rewards support reinforcement learning.
- Supervised learning from labels
- Unsupervised pattern discovery
- Semi-supervised pseudo-labelling
- Reinforcement learning from rewards
- Regression and classification outputs
- Ranking, clustering and anomaly detection
The Business Analytics Lifecycle
Connect problem framing, task translation, data, training, deployment and actionable insight.The course organises AI-driven business analytics into a six-stage loop: understand the business problem, translate it into a machine-learning problem, prepare relevant data, train and test models, deploy a selected model for inference, and generate actionable insight.
- Business problem definition
- Machine-learning translation
- Relevant data preparation
- Training and testing
- Inference deployment
- Actionable insight
Problem Framing and Model Choice
Define the unit, target, timing and action before comparing algorithms and operational constraints.Problem framing turns a broad organisational concern into a decision that can be observed and improved. This chapter uses context, causes, stakeholders and expected outcomes to define the unit, target, timing and action.
- Accountable role and affected group
- 5W1H context framing
- Problem statement and causes
- Unit, target and horizon
- Task and data fit
- Operational constraints
Data Preparation and Feature Engineering
Clean, encode, scale, create and select features without leaking future or test information.Data preparation controls what a model is able to learn.
- Missing values and valid outliers
- One-hot and ordinal encoding
- Min-max and standard scaling
- Log transformation
- Feature creation and selection
- Principal component analysis
Regression and Classification
Interpret numeric forecasts, class probabilities, coefficients, residuals and decision thresholds.Linear regression predicts a numeric outcome with an interpretable fitted relationship, while logistic regression estimates the probability of a binary outcome.
- Continuous numeric targets
- Binary probability outputs
- Linear coefficients and units
- Logistic coefficient signs
- Residual interpretation
- Decision thresholds
Trees, Ensembles and Model Evaluation
Compare readable rules with bagging and boosting, then recompute classification metrics from errors.Decision trees learn readable rules but can be unstable and prone to overfitting. Random forests reduce variance by combining bootstrap trees, while boosting learns sequentially from earlier errors.
- Decision-tree rules
- Tree instability and overfitting
- Random-forest bagging
- Sequential boosting
- Training and unseen testing
- Confusion-matrix cells
Deep Learning for Structured Data
Trace weights, biases and activations through networks, transfer learning and sequence architectures.Neural networks transform inputs through learned weights, biases and activations. This chapter traces one forward pass before scaling to fully connected networks, recurrent networks, LSTMs and transformers.
- Weights and biases
- Activation functions
- Fully connected networks
- RNN and LSTM sequences
- Transformer attention
- Transfer learning
Unstructured Data and Image Inference
Extract features from text and images, preserve preprocessing contracts and govern confidence thresholds.Text, images, audio and video must be transformed into structured representations before models can compare them. This chapter covers n-grams, named entities, sentiment features, convolutional feature maps and dimensionality reduction.
- Text feature extraction
- Image feature mapping
- Dimensionality reduction
- Pretrained CNN inference
- Resize and normalisation
- Batch and label order
Generative, Agentic and Responsible AI
Control generation, tool actions, process evidence, attribution, privacy and human approval.Generative models create content or representations, while agentic workflows add goals, state, plans, tool calls and repeated action. This chapter separates generation quality from tool reliability and policy compliance.
- Transformers and generation
- Autoencoders and GANs
- Agent planning and tools
- Least-privilege actions
- Human approval points
- AI process evidence
The teaching sequence begins with AI foundations, supervised and unsupervised learning, and the main machine-learning task families. It then develops a business analytics process that starts with understanding the problem, translates it into a machine-learning problem, prepares relevant data, trains and tests candidate models, deploys a selected model for inference, and turns findings into actionable insight.
Structured-data work includes cleaning, encoding, scaling, feature creation and selection, linear and logistic regression, decision trees, random forests, neural networks and performance evaluation. Unstructured-data work covers feature extraction from text and images, convolutional neural networks, pretrained image inference, clustering, transformers, autoencoders and generative adversarial networks.
The course is assessed entirely through coursework in the published table: an individual hands-on analytical brief, an individual AI solution proposal, a team proof-of-concept with a pitch and reflection, and weekly quizzes plus seminar engagement. The course also has a detailed generative AI policy.
Assessments 1 to 3 permit assistance with attribution after the student's own initial attempt and require a process evidence appendix; the weekly quiz component is no-assistance.
Across every topic, the central skill is accountable reasoning: define the decision, protect the data boundary, choose a suitable task and model, evaluate errors on unseen evidence, communicate what the result means, and monitor what happens when the organisation acts.
Worked example · free
Choose a threshold from a confusion matrix
- +1Precision = TP / (TP + FP) = 36 / (36 + 12) = 36 / 48 = 0.75, so 75% of flagged cases are truly positive.
- +1Recall = TP / (TP + FN) = 36 / (36 + 9) = 36 / 45 = 0.80, so the model catches 80% of actual positives.
- +1F1 = 2PR / (P + R) = 2 x 0.75 x 0.80 / 1.55 = 0.774, or about 77.4%.
- +1A lower threshold normally flags more cases, tending to increase recall and review workload while risking more false positives and lower precision. The decision depends on review cost and the harm of a miss.
Key terms
- Supervised learning
- Learning a mapping from inputs to a known target using labelled examples, then evaluating that mapping on unseen evidence.
- Unsupervised learning
- Finding structure such as clusters, anomalies or representations in data without supplied target labels.
- Regression
- A supervised task that predicts a numeric value, such as demand, spend or time.
- Classification
- A supervised task that predicts a category or estimates the probability of a category.
- Feature engineering
- Creating or transforming input variables so they express relationships useful to the modelled task.
- Data leakage
- Information entering training or selection that would not be available when the model makes a real prediction.
- Overfitting
- Learning training-specific noise or detail so closely that performance deteriorates on new observations.
- Confusion matrix
- A table of true positives, false positives, false negatives and true negatives for classification decisions.
- Precision
- The proportion of predicted positive cases that are actually positive.
- Recall
- The proportion of actual positive cases that the model correctly identifies.
- Transfer learning
- Adapting a pretrained model to a new task by replacing or fine-tuning selected layers.
- Feature extraction
- Transforming raw text, images, audio or other inputs into structured numeric representations.
- Generative AI
- Models that produce new content or synthetic data from learned patterns and supplied context.
- Agentic workflow
- A goal-directed loop in which a system plans, calls tools, observes results and chooses subsequent actions.
- AI process evidence
- A reviewable record of tools, purposes, interactions, outputs, evaluation, revision and attribution.
INFS3822 FAQ
How is this course assessed?
The published Term 3 2026 table has four coursework components: a 25% hands-on analytical brief, a 20% AI solution proposal, a 40% team proof-of-concept package, and 15% preparation and participation. The numeric weights total 100%, and every row is marked as not a hurdle.
How demanding is the course?
The difficulty comes from joining technical and business reasoning. You need to prepare data, choose and evaluate models, interpret errors, explain findings to a business audience, and defend the action that follows. The source does not publish a difficulty rating or pass rate, so the useful preparation is repeated end-to-end practice.
Is assessment entirely coursework?
Yes. The official T3 2026 assessment table lists four coursework components only. Preparation should follow each analytical brief, proposal, proof-of-concept, quiz and seminar requirement rather than assuming an additional end-of-term test. Check Moodle for the current task brief and submission details.
Which models are taught?
The course materials cover linear and logistic regression, decision trees, random forests, fully connected neural networks, recurrent networks, transformers, convolutional networks, clustering, autoencoders and generative adversarial networks. Model choice is framed through the business target, data type, performance, operational constraints and interpretability.
Can I use generative AI in assessments?
The rule depends on the component. Assessments 1 to 3 use Level 4, Assistance with Attribution, after your own initial attempt and with a required AI Use and Process Evidence Appendix. Assessment 4 Part A weekly quizzes use Level 1, No Assistance, including no AI-enabled browser.
What should a proof-of-concept demonstrate?
A credible proof-of-concept should connect a defined business problem to prepared evidence, a justified analytical method, evaluation on representative cases, a usable output and a monitored action. For generative or agentic workflows, it should also expose tool permissions, approval points, failure handling, logs and evidence of student judgement.
Questions students ask in each INFS3822 chapter
- Why is a prediction not a decision?
- A prediction estimates an outcome from learned patterns. A decision adds a threshold, error costs, organisational constraints, fairness and a chosen action. Two teams can receive the same probability and act differently because their review capacity and consequences differ. A strong analysis keeps those layers separate. Chapter 01 · AI Foundations and Business Value
- How do I distinguish supervised and unsupervised learning?
- Ask whether every training example has a known target. If it does, the model can learn supervised prediction. If it does not, clustering, anomaly detection or representation learning can discover structure, but a person still must interpret whether that structure supports the business decision. Chapter 02 · Learning Paradigms and ML Tasks
- Why is the lifecycle a loop?
- Deployment changes behaviour, data and sometimes the business process itself. Monitoring can show that a target is unstable, features no longer represent current cases, error costs were misjudged or the action creates little value. Any of those findings can send the team back to an earlier stage. Chapter 03 · The Business Analytics Lifecycle
- What belongs in a problem statement?
- State what is happening, to whom, where and when, why it matters, the likely drivers and the decision that could improve the result. Translate broad terms into observable definitions. The statement should support a target, evidence plan and action rather than simply announcing that AI will be used. Chapter 04 · Problem Framing and Model Choice
- Why must preparation be fitted on training data?
- Imputation values, scaling parameters, selected features and components are learned from data. Letting the test set influence them leaks information from supposedly unseen evidence and produces an optimistic evaluation. Fit on training rows, freeze the transformation and apply it unchanged to testing rows. Chapter 05 · Data Preparation and Feature Engineering
- Why is logistic regression used for classification?
- It transforms a weighted combination of features into a probability between zero and one. That probability can be retained for ranking or compared with a threshold to produce a class. The threshold is chosen from operational and error costs rather than learned as a universal truth. Chapter 06 · Regression and Classification
- Why can accuracy be misleading?
- When one class dominates, a model can predict only the majority class and still report high accuracy. Inspect the confusion matrix and class-specific precision, recall and F1. Compare with a simple majority baseline and connect the missed cases and false alarms to operational consequences. Chapter 07 · Trees, Ensembles and Model Evaluation
- When should I consider a deep network?
- Consider it when the data volume and pattern complexity support representation learning that simpler baselines cannot capture adequately. The proposal should also address computation, latency, interpretability and monitoring. Compare against a credible conventional model before claiming that depth adds value. Chapter 08 · Deep Learning for Structured Data
- Why does preprocessing need to match training?
- A model learns patterns in a particular input representation. Changed channel order, scale, crop, resize or field mapping can move live inputs away from that representation and corrupt predictions. Document the contract and test reference examples whenever the deployment pipeline changes. Chapter 09 · Unstructured Data and Image Inference
- How does agentic AI differ from generation?
- A generative model produces content from context. An agentic workflow maintains a goal or state, selects actions, calls tools, observes results and repeats. Tool use adds permission, validation, logging, recovery and approval requirements that content quality alone cannot satisfy. Chapter 10 · Generative, Agentic and Responsible AI
How to prepare for the assessments
Study the course as a connected decision process. First learn to identify the unit, target, prediction time and action in any case. Next practise routing the required output to regression, classification, ranking, clustering, anomaly detection or recommendation. Build a preparation checklist for missing values, encoding, scaling, feature creation, feature selection and leakage, and apply it before every model.
Recompute confusion-matrix metrics by hand until the precision and recall denominators are automatic. Compare every complex model with a simpler baseline and write one sentence on the operational and governance trade-off. For unstructured data, memorise the full inference contract: input shape, channel order, normalisation, label order and review threshold.
Finally, maintain your AI process evidence as you work rather than reconstructing it at submission time. Keep dated drafts, record why each tool was used, verify suggestions independently, attribute incorporated output and practise explaining every decision aloud. The highest-value revision is a backward audit from recommendation to prediction, metric, test evidence, prepared features and original business goal.
Your AI Computer Science tutor for INFS3822
Stuck on a hard INFS3822 question? Sia is AskSia’s AI Computer Science tutor — ask any INFS3822 Artificial Intelligence for Business Analytics 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.
