👁️ Smart-ITAM-Analytics: Hybrid Risk Scoring & ML Pipeline

A hybrid IT asset risk analytics system that combines rule-based operational alerts with machine learning-based replacement risk scoring. This project builds an end-to-end pipeline from raw maintenance logs to predictive decision support, enabling proactive IT asset management.

---

📌 Inspiration & Background (The Dcard Experience)

This project was inspired by my experience as an IT MIS intern at Dcard, where I managed over 2,000 hardware assets and executed a full-scale physical re-audit following disaster recovery efforts.

I observed that IT operations are often highly reactive — assets are only replaced after failure or when maintenance costs peak. This project bridges IT infrastructure and data science to develop a hybrid system that detects immediate operational risks and predicts future replacement needs.

---

🎯 Project Objectives

• Operational Safety: Identify assets requiring immediate attention via rule-based filters.
• Cost Prediction: Use Machine Learning to identify "High-Cost" assets (Top 25% maintenance expenses).
• Strategic Planning: Generate a Composite Risk Score to prioritize budget allocation.

---

🏗️ System Architecture

graph TD
    A[Raw Maintenance Logs] --> B[Automated ETL Pipeline]
    B --> C[Feature Engineering]
    C --> D{Hybrid Risk Engine}
    D -->|Predictive| E[XGBoost Classifier: Future Cost Risk]
    D -->|Rule-Based| F[Compliance & Warranty Filters: Immediate Risk]
    E --> G[Composite Risk Scoring]
    F --> G
    G --> H[Actionable Replacement Ranking]
    H --> I[Future Roadmap: CI/CD & Alerting]

Loading

---

⚠️ Rule-Based Risk Detection

To capture urgent operational risks that require manual intervention:

📌 Selection Criteria

• Non-Compliant Assets: Devices failing security or OS standards (e.g., outdated patches).
• Warranty Expiry: Assets with warranties expiring within 90 days.

📊 Practical Value

This reflects real-world IT workflows where compliance and hardware support status dictate immediate replacement cycles, regardless of predicted costs.

---

🤖 Machine Learning & Target Formulation

1. Target Definition: Why "High-Cost"?

• Label: Is_High_Cost = 1 if total maintenance expense falls within the Top 25% (Upper Quartile).
• Strategy: Predicting exact dollar amounts is often volatile due to random repair events. Classifying assets into Risk Tiers provides a more stable and actionable signal for procurement planning.

2. Model Optimization (Handling Imbalance)

To address the 3:1 data imbalance (Normal vs. High-Cost), I implemented Cost-Sensitive Learning (scale_pos_weight = 3):

• Trade-off: I deliberately sacrificed 3% of overall Accuracy to achieve a 42% increase in Recall (0.71).
• Why?: In ITAM, a "False Negative" (missing a failing device) is far more expensive than a "False Positive" (inspecting a healthy one).

3. Feature Engineering

• Asset_Age: Years since purchase.
• Repair_Intensity: Total Repair Count / Asset Age.
• Compliance_Score: Categorical encoding of asset health.
• Categorical Data: Asset_Type, Department, Warranty_Status.

---

📊 Data-Driven Insights (EDA)

Findings from eda_visualization.ipynb that shaped the final model:

• Cost Driver: Non-compliant assets contribute to approximately $616k in total maintenance costs, significantly higher than compliant ones.
• Correlation: Repair_Count shows a moderate correlation (~0.47) with cost, making it a strong predictor for the ML model.
• Asset Health: Laptops and Servers in the "At Risk" category represent the highest concentration of potential savings through proactive replacement.

---

⚙️ Replacement Risk Scoring

To provide a single source of truth for IT managers, we calculate a Composite Risk Score (0.0 - 1.0):

$$\text{Risk Score} = 0.4(P_{ML}) + 0.2(\text{Age}_{norm}) + 0.2(\text{Repairs}_{norm}) + 0.2(\text{Warranty}_{exp})$$

Component	Weight	Significance
ML Probability (P_ML)	40%	Captures non-linear failure patterns and hidden risks identified by XGBoost.
Asset Age (Age)	20%	Accounts for the physical law of hardware depreciation and mechanical aging.
Repair Intensity	20%	Reflects historical reliability; frequent repairs indicate a "Lemon" asset.
Warranty Status	20%	Represents financial risk; expired warranties lead to 100% out-of-pocket costs.

Score Range	Risk Level	Action Recommended
0.0 – 0.4	Low	Routine Maintenance
0.4 – 0.7	Medium	Monitor & Budget for Next Year
0.7 – 1.0	High	Immediate Replacement Planning

---

💼 Business Impact

• From Reactive to Proactive: Reduces "firefighting" by predicting failures before they occur.
• Budget Optimization: Scientific justification for hardware procurement based on cost-risk ROI.
• Compliance Security: Integrates IT security (compliance) directly into the asset lifecycle.

---

🛠 Tech Stack & Structure

• Languages: Python (Pandas, NumPy)

• Machine Learning: XGBoost (Cost-Sensitive Learning), Scikit-learn

• Visualization: Matplotlib, Seaborn

• Data Engineering: SQL, ETL Pipelines

• Project Structure:

.
├── data/
│   ├── raw_it_assets.csv          # Original maintenance logs
│   ├── processed_it_assets.csv    # Cleaned data after ETL
│   └── final_risk_assessment.csv  # Output with Risk Scores & Levels
├── notebooks/
│   ├── data_cleaning_ETL.ipynb    # Preprocessing & Feature Engineering
│   ├── eda_visualization.ipynb    # Statistical & Cost Analysis
│   └── predictive_model.ipynb     # XGBoost & Scoring Engine
├── .gitignore                     # To exclude .venv and large data files
└── README.md

---

🚀 Future Roadmap: CI/CD & Automated Alerting

• Data CI/CD & Alerting: Integrate GitHub Actions to automate daily scoring and trigger Slack/Email alerts when an asset's risk escalates.
• Risk Escalation Detection (Delta): Develop a monitoring module that compares T vs T-1 risk scores to identify assets jumping from "Medium" to "High" risk levels.
• Time-Series Integration: Forecast specific month-of-failure for server-grade hardware.
• Interactive Dashboard: Build a Streamlit UI for real-time risk exploration.

---

📝 Final Note

This project demonstrates the ability to translate technical data science workflows into practical IT solutions, grounded in the reality of managing large-scale enterprise infrastructure.