Back to Project Library
Completed
Sanitized proof

Adidas IT · Gradient Boosting

Ticket Reassignment Prediction

Modeled IT ticket reassignment behavior to reduce escalation friction, rerouting cost, and downtime risk.

Pythonscikit-learnFeature EngineeringModel Evaluation

Model Confidence Distribution

Accuracy76%
Recall (Critical Catch Rate)86%
F1 Score73%

Academic scenario: approximately $277K in modeled annual labor savings under assumed ticket costs and benefits, not realized savings.

Ticket Reassignment Prediction

Adidas IT · Gradient Boosting

What this proves

  • Problem: IT support workflows were losing time and retail uptime through preventable ticket reassignments.
  • Method: built a Gradient Boosting classifier around urgency, service, category, and geography variables, then evaluated it with accuracy, recall, and F1.
  • Outcome: produced a predictive routing aid with approximately 76.1% accuracy, 85.8% recall, 73.3% F1, and about $277K in modeled annual labor savings (not realized savings) cited in the project materials.

Contribution, validation, and limits

This was an academic team project, not a deployed cost-saving system. Michael contributed as part of a six-person team; the presentation reports classifier performance and a proposed routing workflow. The deck is the available evidence, not a reproducible production benchmark.

The approximately $277K annual labor-savings scenario assumes about $1,000 benefit for each correctly predicted reassigned ticket and $833 cost for each false positive. These are modeling assumptions, not realized savings. The deck's separate retail-downtime estimate uses a $10,000/day store-sales assumption explicitly attributed to ChatGPT and is not presented here as business impact.

The deck describes oversampling in training but its code outline lists oversampling before the train/test split. The exact pipeline must be inspected before claiming leakage-free validation. Reported accuracy, recall, and F1 are presentation results, not an independently reproduced evaluation.

Live proof surface

The public-safe proof for this project is the final presentation deck. It shows the problem framing, feature set, evaluation logic, and business-facing recommendation path without exposing private ticket data.

Open presentation deck

Proof note

The live deck is the strongest public artifact for this project. The raw dataset, feature engineering notebook, and internal workflow details remain sanitized.