Practical driving exams currently rely primarily on trained examiner judgement, despite documented variability in examiner agreement. No objective data layer currently supports examiner review at the level of specific moments during the drive, because examiners cannot annotate behaviour while attending to the candidate and the road. Existing telemetry-based approaches have mostly relied on whole-drive averages or uniform acceleration and jerk thresholds, which do not differentiate behaviour by road context. 

This thesis examined how raw GPS telemetry could be transformed into behavioural information, how road context could be reconstructed from the same data, and how these two layers could be combined into examiner-readable behavioural evidence. After filtering, the dataset comprised 113 simulated driving examinations. Unsupervised clustering of 5-second telemetry windows produced six behavioural states (i.e. braking, near-stop, smooth urban cruising, high-speed stable cruising, high-volatility, and acceleration). GPS curvature combined with OpenStreetMap information produced four road context categories (i.e. straight driving, simple turns, sustained turns, and roundabouts). 

The two layers were combined into 5,704 event-level behavioural profiles, each describing the behavioural sequence within one road-context event (e.g. braking → high-volatility within a roundabout) and scored against comparable events in the same context and speed zone. Behaviour was strongly context-dependent: high-volatility behaviour was nearly eleven times more common in sustained turns than during straight driving. Weak-label validation against examiner outcomes suggested higher strongest-event anomaly scores in more severe outcome groups, but bootstrap intervals were consistent with chance. The findings support a decision-support interpretation: telemetry can enrich route visualisation through dashboard-ready behavioural anomaly points that show what happened at specific road-demand moments and why those moments stood out, while leaving assessment interpretation to the examiner.