COMP90089 Machine Learning Applications for Health
The Complete Study & Assessment Guide · S2 2026COMP90089 Overview
- Faculty of Engineering and Information Technology
- Semester 2, 2026
- a postgraduate 12.5-point subject
- 12.5 credit points
- a clinical machine-learning applications subject
COMP90089 covers health systems, clinical data, privacy, digital phenotyping, diagnostic reasoning, task and study design, accuracy, AI lifecycle, deep learning, NLP, LLMs, agents, implementation and regulation. It is taught within Faculty of Engineering and Information Technology. It is a postgraduate 12.5-point subject.
- Prediction time is a firewall Every feature must be available before the decision the model is meant to support.
- Accuracy needs context Prevalence, error costs, calibration and workflow determine whether a metric is clinically useful.
- Data encode care Missingness, labels and documentation reflect health-system processes as well as patient state.
What you get in the full COMP90089 Study & Assessment Guide
- 9 chapters, 25 pages — every assessed topic, week by week
- 12 original practice questions, each with a model answer
- 9 worked examples, solved the way COMP90089 marks
- 19 computed figures, redrawn from the course’s own numbers
- Before-you-submit checklist for the Group report and presentation (40%) — what markers reward, what costs marks
Free on this page: assessment table, chapter map, one worked example, key terms, FAQ. The rest is in the full study & assessment guide.
How COMP90089 is assessed
| Component | Weight | Format |
|---|---|---|
| Weekly quizzes | 20% | Weeks 2–11 |
| Two individual programming assignments · hurdle | 30% | Weeks 4 and 10; collective 50% pass threshold |
| Group project proposal | 10% | Oral proposal in Week 8 |
| Group report and presentation · hurdle | 40% | 20% report plus 20% oral presentation; collective 50% group-assignment threshold |
Current assessment is 20% weekly quizzes, 30% individual programming assignments, a 10% group proposal and a 40% group report/presentation. Students need at least 50% across each assignment group.
What other students are asking about COMP90089
- Where can I find lecture notes or study guides for COMP90089 at UniMelb?
- The COMP90089 study guide in AskSia Library covers Machine Learning Applications for Health in 9 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 COMP90089?
- No final exam is listed in the offering this AskSia Library guide is built from; COMP90089 is assessed by Weekly quizzes 20%, Two individual programming assignments 30%, Group project proposal 10%, Group report and presentation 40%.
- What does COMP90089 cover on clinical data sources, types and missingness?
- In AskSia Library's COMP90089 guide, the Clinical Data Sources, Types and Missingness chapter explains electronic health record · clinical data type · informative missingness · select and align data types while preserving time, unit and care-process meaning.
- What does COMP90089 cover on health-data governance?
- In AskSia Library's COMP90089 guide, the Health-data Governance chapter explains de-identification · data governance · consent · design a health-data project with proportionate access and re-identification controls.
- What does COMP90089 cover on digital phenotyping and longitudinal signals?
- In AskSia Library's COMP90089 guide, the Digital Phenotyping and Longitudinal Signals chapter explains digital phenotype · sampling frequency · concept drift · engineer longitudinal features without confusing device behaviour with patient state.
Current COMP90089 dates
| Date | Item | Control |
|---|---|---|
| Week 4 | Individual programming assignment 1 | Part of the 30% individual-assignment group and its collective threshold. |
| Week 8 | Group project proposal | 10% oral presentation. |
| 3 November 2026 | Project presentation week | Specific schedule is set by the teaching team. |
Dates are as published in the current course materials. Confirm exact deadlines and submission settings in the live LMS.
What COMP90089 covers
Read Health Informatics as the foundation, Diagnostic Reasoning and Bayesian Updating as the main change in method, and Implementation, Evidence to Decision and Regulation as the final application of the course.
Health Informatics
biomedical informatics · learning health system · feedback loop · map how clinical data could produce and evaluate a care improvementDefine biomedical informatics
The course material gives this chapter a concrete anchor: The opening lectures define biomedical informatics and the learning-health-system feedback cycle.
- Biomedical informatics
- Learning health system
- Feedback loop
- Applying biomedical informatics
- Limits of learning health system and feedback loop
Clinical Data Sources, Types and Missingness
electronic health record · clinical data type · informative missingness · select and align data types while preserving time, unit and care-process meaningDefine electronic health record
The course material gives this chapter a concrete anchor: Lecture 4 details clinical data types; MIMIC access pages expose practical table and governance constraints.
- Electronic health record
- Clinical data type
- Informative missingness
- Applying electronic health record
- Limits of clinical data type and informative missingness
Health-data Governance
de-identification · data governance · consent · design a health-data project with proportionate access and re-identification controlsDefine de-identification
The course material gives this chapter a concrete anchor: Week 3 explicitly covers privacy, de-identification, ethics and governance alongside national ethical guidance.
- De-identification
- Data governance
- Consent
- Applying de-identification
- Limits of data governance and consent
Digital Phenotyping and Longitudinal Signals
digital phenotype · sampling frequency · concept drift · engineer longitudinal features without confusing device behaviour with patient stateDefine digital phenotype
The course material gives this chapter a concrete anchor: Weeks 4 and 5 place digital phenotyping before diagnostic reasoning and task definition.
- Digital phenotype
- Sampling frequency
- Concept drift
- Applying digital phenotype
- Limits of sampling frequency and concept drift
Diagnostic Reasoning and Bayesian Updating
pre-test probability · likelihood ratio · post-test probability · translate a model or test result into updated clinical probabilityDefine pre-test probability
The course material gives this chapter a concrete anchor: Weeks 5–6 make diagnostic reasoning and clinical task definition central.
- Pre-test probability
- Likelihood ratio
- Post-test probability
- Applying pre-test probability
- Limits of likelihood ratio and post-test probability
Clinical Task Definition and Study Design
index time · target label · data leakage · define cohort, comparator, prediction horizon and validation before modellingDefine index time
The course material gives this chapter a concrete anchor: Week 6 pairs defining the clinical task with study design.
- Index time
- Target label
- Data leakage
- Applying index time
- Limits of target label and data leakage
Clinical Accuracy, Calibration and Thresholds
sensitivity · positive predictive value · calibration · calculate confusion-matrix metrics and select a threshold from consequencesDefine sensitivity
The course material gives this chapter a concrete anchor: Week 7 directly covers clinical accuracy metrics and the AI lifecycle.
- Sensitivity
- Positive predictive value
- Calibration
- Applying sensitivity
- Limits of positive predictive value and calibration
Deep Learning, Clinical NLP and Language Models
representation learning · clinical natural language processing · large language model · select and evaluate a deep or language model for a specific clinical information taskDefine representation learning
The course material gives this chapter a concrete anchor: Weeks 8–10 cover deep learning, clinical NLP, LLMs, agents and combined methods.
- Representation learning
- Clinical natural language processing
- Large language model
- Applying representation learning
- Limits of clinical natural language processing and large language model
Implementation, Evidence to Decision and Regulation
workflow integration · external validation · post-deployment monitoring · move from retrospective model evidence to a monitored care interventionDefine workflow integration
The course material gives this chapter a concrete anchor: Weeks 8 and 11 join workflow implementation, evidence-to-decision and legal or regulatory issues.
- Workflow integration
- External validation
- Post-deployment monitoring
- Applying workflow integration
- Limits of external validation and post-deployment monitoring
It carries 12.5 credit points.
It is positioned as a clinical machine-learning applications subject.
The subject centres clinical task definition and implementation rather than treating health records as a generic benchmark dataset.
Assessment in COMP90089 is distributed as follows: 20% weekly quizzes, 30% individual programming assignments, a 10% group proposal and a 40% group report/presentation
The operational assessment conditions matter here.
The subject culminates in the group report and oral presentation.
What makes COMP90089 demanding is concrete: turning a convenient retrospective prediction target into a clinically meaningful, leakage-free and prospectively implementable decision with honest uncertainty
Students must achieve at least 50% across the individual assignments and at least 50% across the group assignments, as well as the ordinary overall subject pass.
For enrolment planning, Consult the current official enrolment and eligibility controls for this offering.
Read Health Informatics as the foundation, Diagnostic Reasoning and Bayesian Updating as the main change in method, and Implementation, Evidence to Decision and Regulation as the final application of the course.
Worked example · free
Detect outcome leakage
- 1Fix prediction time at discharge.
- 1Draw a timestamp for every feature.
- 1Remove the post-discharge code from development and evaluation.
- 1Rebuild the comparator and metrics.
- 1Explain prospective data-availability testing.
Key terms
- Biomedical informatics
- Interdisciplinary study of data, information and knowledge in health and biomedicine. This chapter uses the concept when students map how clinical data could produce and evaluate a care improvement.
- Learning health system
- Health system that systematically turns care data into evidence and evidence into improved care. It helps explain the reasoning required to map how clinical data could produce and evaluate a care improvement.
- Feedback loop
- Cycle in which observed outcomes inform a changed intervention and later measurement. Its limit matters because routine data do not create learning without action, outcome measurement and governance.
- Electronic health record
- Longitudinal digital record produced through care, administration and documentation processes. This chapter uses the concept when students select and align data types while preserving time, unit and care-process meaning.
- Clinical data type
- Representation such as coded event, laboratory value, note, image, waveform or medication order. It helps explain the reasoning required to select and align data types while preserving time, unit and care-process meaning.
- Informative missingness
- Absence whose probability depends on patient state or care process and can carry signal or bias. Its limit matters because a missing test can mean not needed, not ordered, inaccessible or not recorded.
- De-identification
- Transformation and control intended to reduce the chance data can be linked to an individual. This chapter uses the concept when students design a health-data project with proportionate access and re-identification controls.
- Data governance
- Accountability, rules and processes controlling data access, quality, use and stewardship. It helps explain the reasoning required to design a health-data project with proportionate access and re-identification controls.
- Consent
- Context-specific, informed and voluntary agreement to a defined data or research activity. Its limit matters because removing names does not eliminate linkage, group harm or inappropriate purpose.
- Digital phenotype
- Quantified pattern derived from personal digital interaction or sensor data and linked to health-related behaviour or state. This chapter uses the concept when students engineer longitudinal features without confusing device behaviour with patient state.
- Sampling frequency
- Number or timing of observations collected per unit time. It helps explain the reasoning required to engineer longitudinal features without confusing device behaviour with patient state.
- Concept drift
- Change over time in the relationship between inputs, target and deployment context. Its limit matters because sensor availability, platform change and surveillance burden can dominate the apparent health signal.
- Pre-test probability
- Estimated chance of a condition before receiving the current test result. This chapter uses the concept when students translate a model or test result into updated clinical probability.
- Likelihood ratio
- Ratio comparing how probable a test result is with versus without the condition. It helps explain the reasoning required to translate a model or test result into updated clinical probability.
COMP90089 FAQ
Where is the hardest reasoning in Machine Learning Applications for Health?
Turning a convenient retrospective prediction target into a clinically meaningful, leakage-free and prospectively implementable decision with honest uncertainty. COMP90089 covers health systems, clinical data, privacy, digital phenotyping, diagnostic reasoning, task and study design, accuracy, AI lifecycle, deep learning, NLP, LLMs, agents, implementation and regulation.
How does assessment work in Machine Learning Applications for Health?
20% weekly quizzes, 30% individual programming assignments, a 10% group proposal and a 40% group report/presentation. The subject culminates in the group report and oral presentation. Students must achieve at least 50% across the individual assignments and at least 50% across the group assignments, as well as the ordinary overall subject pass.
What form does the final assessed task take in Machine Learning Applications for Health?
The subject culminates in the group report and oral presentation. 20% weekly quizzes, 30% individual programming assignments, a 10% group proposal and a 40% group report/presentation. Students must achieve at least 50% across the individual assignments and at least 50% across the group assignments, as well as the ordinary overall subject pass.
Which pass conditions apply in Machine Learning Applications for Health?
Students must achieve at least 50% across the individual assignments and at least 50% across the group assignments, as well as the ordinary overall subject pass. 20% weekly quizzes, 30% individual programming assignments, a 10% group proposal and a 40% group report/presentation.
Which teaching period does this Machine Learning Applications for Health resource cover?
It is aligned to Semester 2, 2026; confirm your enrolled class and timetable in the current institutional system. Individual programming assignment 1: Week 4; Group project proposal: Week 8; Project presentation week: 3 November 2026.
What should a student check before enrolling in Machine Learning Applications for Health?
Consult the current official enrolment and eligibility controls for this offering. This resource covers Semester 2, 2026. COMP90089 covers health systems, clinical data, privacy, digital phenotyping, diagnostic reasoning, task and study design, accuracy, AI lifecycle, deep learning, NLP, LLMs, agents, implementation and regulation.
Who controls the official rules for Machine Learning Applications for Health?
The university does. This is an independent COMP90089 study resource; current institutional instructions remain authoritative for assessment operation. COMP90089 covers health systems, clinical data, privacy, digital phenotyping, diagnostic reasoning, task and study design, accuracy, AI lifecycle, deep learning, NLP, LLMs, agents, implementation and regulation.
Questions students ask in each COMP90089 chapter
- What belongs in the structure used to map how clinical data could produce and evaluate a care improvement?
- Map how clinical data could produce and evaluate a care improvement. The opening lectures define biomedical informatics and the learning-health-system feedback cycle. Interdisciplinary study of data, information and knowledge in health and biomedicine. This chapter uses the concept when students map how clinical data could produce and evaluate a care improvement. Chapter 01 · Health Informatics
- Which criteria should govern an attempt to select and align data types while preserving time, unit and care-process meaning?
- Select and align data types while preserving time, unit and care-process meaning. Lecture 4 details clinical data types; MIMIC access pages expose practical table and governance constraints. Chapter 02 · Clinical Data Sources, Types and Missingness
- Which constraints shape the work needed to design a health-data project with proportionate access and re-identification controls?
- Design a health-data project with proportionate access and re-identification controls. Week 3 explicitly covers privacy, de-identification, ethics and governance alongside national ethical guidance. Transformation and control intended to reduce the chance data can be linked to an individual. This chapter uses the concept when students design a health-data project with proportionate access and re-identification controls. Chapter 03 · Health-data Governance
- Which constraints shape the work needed to engineer longitudinal features without confusing device behaviour with patient state?
- Engineer longitudinal features without confusing device behaviour with patient state. Weeks 4 and 5 place digital phenotyping before diagnostic reasoning and task definition. Quantified pattern derived from personal digital interaction or sensor data and linked to health-related behaviour or state. This chapter uses the concept when students engineer longitudinal features without confusing device behaviour with patient state. Chapter 04 · Digital Phenotyping and Longitudinal Signals
- What must survive the move required to translate a model or test result into updated clinical probability?
- Translate a model or test result into updated clinical probability. Weeks 5–6 make diagnostic reasoning and clinical task definition central. Estimated chance of a condition before receiving the current test result. This chapter uses the concept when students translate a model or test result into updated clinical probability. Chapter 05 · Diagnostic Reasoning and Bayesian Updating
- What must be fixed before students define cohort, comparator, prediction horizon and validation before modelling?
- Define cohort, comparator, prediction horizon and validation before modelling. Week 6 pairs defining the clinical task with study design. Timestamp at which prediction inputs close and the intended decision is made. This chapter uses the concept when students define cohort, comparator, prediction horizon and validation before modelling. Chapter 06 · Clinical Task Definition and Study Design
- Which inputs and assumptions control the attempt to calculate confusion-matrix metrics and select a threshold from consequences?
- Calculate confusion-matrix metrics and select a threshold from consequences. Week 7 directly covers clinical accuracy metrics and the AI lifecycle. Proportion of actual positive cases correctly classified positive at a threshold. This chapter uses the concept when students calculate confusion-matrix metrics and select a threshold from consequences. Chapter 07 · Clinical Accuracy, Calibration and Thresholds
- Which criteria should govern an attempt to select and evaluate a deep or language model for a specific clinical information task?
- Select and evaluate a deep or language model for a specific clinical information task. Weeks 8–10 cover deep learning, clinical NLP, LLMs, agents and combined methods. Chapter 08 · Deep Learning, Clinical NLP and Language Models
- What should a student check while trying to move from retrospective model evidence to a monitored care intervention?
- Move from retrospective model evidence to a monitored care intervention. Weeks 8 and 11 join workflow implementation, evidence-to-decision and legal or regulatory issues. Placement of model output, action and accountability inside real clinical work. This chapter uses the concept when students move from retrospective model evidence to a monitored care intervention. Chapter 09 · Implementation, Evidence to Decision and Regulation
How to prepare for the assessments
Retrieve the course map, practise the recurring method—define the clinical decision and prediction time, build the cohort and labels without leakage, select and evaluate a model against a comparator, then judge calibration, subgroup effects, workflow value, ethics and regulation—on changed scenarios, and verify every operational assessment detail in the live institutional system.
Your AI Computer Science tutor for COMP90089
Stuck on a hard COMP90089 question? Sia is AskSia’s AI Computer Science tutor — ask any COMP90089 Machine Learning Applications for Health 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.