DS 310 · Year 3 · Semester 2 · 3 credits · AI & Machine Learning

Data Science Lab for Actuaries (Exam PA / ATPA studio)

From raw claim records to an executive-ready model audit: mastering predictive modelling, exposure mechanics, and communication for Exam PA and ATPA.

Full practice exam · 30 Q · 120 min

Season 1 · 8 episodes

The Commercial Auto Turnaround

When Apex Mutual's flagship commercial auto line bleeds forty-two million dollars from severe adverse selection, newly appointed predictive actuary Maya Lin must rebuild the pricing architecture from raw audit logs to an auditable, regulatory-ready machine learning pipeline before the insurance commissioner's filing deadline.

Protagonist · Maya Lin, ASA, newly promoted Senior Predictive Actuary at Apex Mutual

Setting · Apex Commercial Mutual headquarters in Chicago and the Illinois Department of Insurance rate review hearing room

Stakes · A thirty-five million dollar underwriting loss threatening Apex's A.M. Best financial strength rating, ninety thousand policyholders facing cancellation, and an impending rate freeze from the state regulator.

Pick an episode

Episodes

Season finale: Maya defends a leak-free, exposure-weighted hybrid model before the state insurance commissioner and Apex's board, turning a projected fourteen percent underwriting deficit into an eight percent rate-adequate surplus with total ASOP 56 compliance.

From the lecturer: Welcome to DS 310. This is not a passive lecture course; it is an active modeling studio designed to prepare you for SOA Exam PA, Exam ATPA, and real-world actuarial data science. When watching these lecture films, keep an active R or Python session open on one side of your screen and an actuarial computation notebook on the other. For every model derivation, write out the log-likelihood, identify the exposure offset or case weight, and state the underlying distribution link explicitly before letting the algorithm optimize it. After every bake-off lecture, practice translating the mathematical metrics—deviance, Gini index, calibration error—into plain business consequences suitable for a Chief Actuary or underwriting committee. SOA Exam PA and ATPA evaluate not just whether your model runs, but whether you can defend your feature choices, justify tree depth versus GLM interpretability, identify data leakage, and communicate technical risk transparently. Treat each module as a timed mini-project: frame the business problem, write reproducible code, audit the output using partial dependence and SHAP values, and draft a structured executive summary. Let us build models that are both statistically sound and commercially actionable.

Cold open

Maya walks into Conference Room 4B at 7:15 AM to find Chief Actuary Dan Vance staring at a red heat map. Apex's commercial auto book ran an eighty-four percent loss ratio in urban delivery fleets last quarter, but the legacy rating algorithm had projected forty-two percent. Dan slides a USB drive across the table: five hundred thousand raw telematics and claim records from policy years 2021 through 2023, due for a complete refiling in forty-five days.

preparing lecture…

Transcript

Translate a business or actuarial objective into a well-defined supervised learning task with appropriate target and exposure variables.
Apply Actuarial Standard of Practice No. 23 to audit raw insurance data for anomalies, inconsistencies, and truncation.
Detect and eliminate subtle forms of target leakage before model training.