Crew Fatigue & Illness Predictor

I want to build a machine-learning system that flags crew members who are likely to turn up fatigued or on the verge of sickness before a shift begins. For every person I can supply two data streams: their detailed work schedules (hours on duty, rest intervals, overtime) and their daily wellness check-ins (self-reported symptoms, health scores, notes).

Your model should fuse both streams, learn the patterns that precede absence or under-performance, and return a single warning message per crew member. The message must name the primary factors driving the alert—think short rest, long run of night shifts, or a cluster of cold-like symptoms—so supervisors instantly know why the risk is high.

Here’s what I need from you:

• Python pipelines that clean, merge, and engineer features from the scheduling and wellness data.
• A supervised model (tree-based boosting, random forest, or deep-learning if you can justify the gain) trained with solid cross-validation and tuned to minimise false negatives.
• Explainability via SHAP, feature importance, or a similar technique so the warning message can cite the top contributors.
• An inference script or lightweight REST API that, given a crew ID and upcoming shift, returns the composed warning.
• A concise README covering setup, model refresh, and integration touch-points with our existing scheduling tool.

In your proposal, outline the algorithms you favour, any extra data you’d request, and how you will evaluate accuracy and robustness. Once delivered, I should be able to plug the model into our environment and start receiving clear, reason-based alerts long before fatigue or illness affect the floor.