IFoA SA10 study plan

Map the Domain 1 syllabus into a one-page concept sheet covering data structures, data quality, and the data lifecycle, writing each item in your own words.

Explain, in a 200-word note, the difference between structured, semi-structured and unstructured data, giving one insurance example of each.

Practise writing SQL against a small practice dataset: select, filter, join two tables, and aggregate with GROUP BY, then annotate each query with what it answers.

List the main data-quality dimensions (completeness, accuracy, consistency, timeliness, validity) and describe one concrete failure mode for each in an insurance context.

Build a small exploratory analysis on an open dataset (for example motor claims or mortality counts) and record summary statistics, missingness and outliers.

Produce and interpret at least four plots (histogram, box plot, scatter, time series) and write one sentence per plot on what it implies for modelling.

Describe how data science is used across at least four actuarial work areas: pricing, reserving, capital, underwriting and claims triage.

Draft a short data-preparation checklist you would apply before any modelling exercise.

Derive the ordinary least squares estimator and state its assumptions, explaining what each assumption means in practical terms.

Contrast ridge and lasso regularisation: write out the penalty terms and explain when each is preferred.

Fit a linear model and a regularised model on the same dataset and compare them using cross-validation error.

Explain bias-variance trade-off with a worked example of underfitting versus overfitting.

Derive the logistic regression model from the log-odds formulation and interpret coefficients as odds ratios.

Implement and interpret a confusion matrix, ROC curve and AUC on a binary classification problem.

Explain how decision trees split data using impurity measures, and describe the pruning and depth-control options that manage overfitting.

Compare a GLM with a tree-based classifier on the same data and justify which you would deploy and why.

Describe bagging, random forests and gradient boosting, stating how each reduces variance or bias.

Explain how cross-validation, train/validation/test splits and hyperparameter tuning fit together in a defensible workflow.

Apply at least one feature-importance or model-interpretation technique and critique its limitations.

Summarise the main clustering and dimension-reduction methods and one actuarial use case for each.

Map a full model lifecycle, from problem framing through data collection, development, validation, deployment and monitoring, and identify the actuarial control points at each stage.

Explain model risk management expectations for data-driven models, including documentation, independent review and ongoing performance monitoring.

Identify the sources of bias that can enter a model through data, target definition and feature choice, and propose a mitigation for each.

Describe how you would communicate model limitations and uncertainty to a non-technical board or committee.

Summarise the professional and ethical obligations that apply when using data and models, including competence, honesty and acting with integrity, in your own words.

Analyse at least three ethical dilemmas in data science (for example proxy discrimination, opaque models, misuse of personal data) using a structured framework.

Explain data protection principles and how they constrain data collection, retention and use in insurance.

Draft a short response to a scenario where a model performs well statistically but has an unfair impact on a group of policyholders.

Sit at least two full timed mock papers under exam conditions and mark them against the command words used.

Review every wrong or incomplete answer and write a correction note for each, linking it to the relevant syllabus area.

Rehearse concise written explanations of five core methods without notes, aiming for clarity and correct terminology.

Consolidate your glossary, formula sheet and ethics frameworks into a single revision pack you can scan in 20 minutes.