Here is the professional-grade, refined prompt. You can copy and paste this directly into your AI interface.

Prompt: Enterprise-Grade Anomaly Detection System for Fraud Prevention

🎭 Role

You are a Senior Data Scientist and Fraud Risk Specialist with expertise in unsupervised machine learning, financial crime detection, and model deployment. You focus on building robust, explainable, and production-ready systems that balance high detection rates with low false-positive operational overhead.

🌐 Context

We are developing a fraud detection engine to identify [TARGET_TRANSACTION_TYPE, e.g., credit card transactions] within a high-volume financial environment. The system must utilize an Isolation Forest approach to identify outliers in unlabeled transaction data. The goal is to move beyond simple thresholds and implement a data-driven model that can adapt to evolving fraud patterns.

🛠️ Task Instruction

Please architect the technical implementation plan and provide the necessary Python/Scikit-Learn code snippets for the following phases:

1. Feature Engineering: Define logic to transform raw transactional data into high-value features, specifically:
   • Temporal features (e.g., cyclical encoding of time of day).
   • Spatial features (e.g., Haversine distance from the previous transaction).
   • Statistical features (e.g., rolling averages/standard deviations of [USER_METRIC]).
2. Model Configuration: Implement an IsolationForest model. Explain the logic for setting the contamination parameter based on historical fraud benchmarks.
3. Anomaly Scoring & Thresholding: Write code to generate anomaly scores and flag the top [PERCENTILE]% of records as suspicious.
4. Performance & Evaluation: Define a methodology to balance precision vs. recall. Propose metrics to track the model’s performance in a production environment to prevent "alert fatigue."
5. Visualization: Provide code to generate a scatter plot showing the data distribution, highlighting the decision boundary, and color-coding flagged anomalies.

⚖️ Constraints & Tone

• Tone: Professional, analytical, and highly technical.
• Best Practices: Prioritize code efficiency, scalability, and anti-leakage techniques (e.g., proper train-test splitting).
• Avoid: Over-simplified explanations; assume the user is proficient in Python and machine learning fundamentals.
• Length: Provide concise explanations followed by clean, commented code blocks.

📝 Output Format

• Executive Summary: A brief overview of the strategy.
• Technical Implementation:
  • Step-by-step breakdown with Markdown headers.
  • Code snippets using pandas and scikit-learn.
• Optimization Strategy: A section specifically addressing how to tune the model to minimize False Positives.
• Visualization: Code for a matplotlib/seaborn visualization.

Placeholders

• [TARGET_TRANSACTION_TYPE]: The specific type of data (e.g., E-commerce payments, B2B wire transfers).
• [USER_METRIC]: The primary variable to monitor (e.g., spending velocity, account balance).
• [PERCENTILE]: The target threshold (e.g., 1%).

Instruction to the User: Before pasting this into the AI, replace the [BRACKETED] variables with your specific project details to get a tailored technical response.