QuantumData holds tables for actuaries who are practicing predictive analytics in the form Exam PA expects. Each object is a rectangular data frame with a documented response, a mix of numeric and categorical predictors, and a row count that still fits in memory on a laptop. The files come from SOA exam sittings, SOA sample projects, and the practice exams written around that syllabus, plus a few public modeling sets used the same way.

The exam task is to explore a file, choose a GLM or a tree, hold out data, and say what the fit means for a decision. These tables are the raw material for that loop. Targets include claim amounts, retention, a high-value flag, hospital days, crash severity, and hourly counts. Predictors are the ones a pricing, underwriting, or health analyst would actually be handed: demographics, prior utilization, road and weather conditions, and a handful of scores built by someone else.

The level is intermediate. A first course in GLMs is enough to start. The practice is in the exam habits around the model: a split that respects time when the rows are ordered, a check that each predictor is known before the outcome, and a short recommendation a claims or pricing lead could use. The UCI bank file is the clearest case. Call duration dominates a model of whether the customer subscribed, and duration is known only after the call ends, so a score meant to be used before the call leaves that column out.

Column dictionaries live in this repository, one roxygen file per table under R/ and the generated help under man/. Several source extracts are also stored as CSV under raw-data/. The packaged .RData objects are not in the git tree. The project notes record that those files were moved to Hugging Face.

Tables

• Exam sittings. june_pa is the June 2019 crash file: 23,137 crashes, a severity score, and road, weather, light, and time features. customer_value is the December 2019 file: 48,842 prospects and a high or low value flag, with an insurance score, capital gains, and hours worked. The June 2020 files are patient_length_of_stay (days in hospital, admission type, and medication indicators, 10,000 stays), customer_phone_calls, and patient_num_labs.
• SOA sample projects. readmission is the 2019 hospital readmission sample: 66,782 stays, a binary readmission flag, prior emergency visits, length of stay, and an HCC risk score. student_success is the 2019 student sample: 585 rows and a wide set of school, family, and study variables.
• Practice-exam files. health_insurance is 1,338 policies with age, BMI, smoking status, and annual charges. apartment_apps is 1,430 buildings with application counts and unit characteristics. exam_pa_titanic is 906 passengers with survival, ticket class, and age.
• Related modeling tables. auto_claim is 10,296 auto policies with claim frequency, claim amount, and whether the policy was retained. bank_loans is the UCI bank marketing extract: 41,188 contacts and 21 variables, including credit default, housing, and campaign history. actuary_salaries is a small DW Simpson table of salary bands by industry, exams, and years of experience. travel_insurance and travel_spending are trip-cost tables. bike_sharing_demand and pedestrian_activity are hourly counts with weather. boston is the 506-tract housing file.