Data & Model Management in DSS: Types, Architecture & Workflows

TAKEAWAYS:

• Four DSS types differ by how they process data/models: data-driven (e.g., eSewa transaction analytics) vs. model-driven (e.g., Ncell network optimization) vs. knowledge-driven (e.g., medical diagnosis systems) vs. document-driven (e.g., legal case analysis).
• Data management in DSS involves internal sources (ERP, TPS) and external sources (government databases, market trends), with data warehouses as the central repository for analysis.
• Model management enables what-if analysis (e.g., "What if NTC increases electricity tariffs by 10%?") and optimization (e.g., Daraz’s delivery route planning) using mathematical models.
• Data-driven DSS (e.g., Khalti’s fraud detection) relies on OLAP cubes and data mining, while model-driven DSS (e.g., Pathao’s surge pricing) uses simulation and optimization.
• Knowledge-driven DSS (e.g., agricultural advisory systems) combines rules, fuzzy logic, and case-based reasoning to mimic expert judgment.
• Document-driven DSS (e.g., legal research tools) processes unstructured text (PDFs, emails) using NLP and text mining.

1. What is Data and Model Management in DSS?

Decision Support Systems (DSS) rely on two core components:

1. Data Management: Collecting, storing, and processing data from multiple sources.
2. Model Management: Applying mathematical, statistical, or AI models to analyze data and generate insights.

Why it matters: DSS cannot function without clean, structured data and reliable models. Poor data quality leads to wrong decisions (e.g., a bank approving a risky loan), while weak models fail to provide actionable insights (e.g., a retail DSS missing sales trends).

2. Types of DSS Based on Data/Model Usage

DSS are classified into four types, each with distinct data/model requirements:

3. Data Management in DSS

A. Data Sources

DSS data comes from two main categories:

1. Internal Data Sources (within the organization):

   • Transaction Processing Systems (TPS): Sales, inventory, customer records (e.g., Daraz’s order history).
   • Enterprise Resource Planning (ERP): Finance, HR, supply chain (e.g., NTC’s billing system).
   • Legacy Systems: Older databases (e.g., bank core banking systems).

2. External Data Sources (outside the organization):

   • Government Databases: NEPSE stock prices, NPR inflation rates.
   • Market Data: Competitor pricing (e.g., Daraz vs. Amazon India).
   • Social Media: Customer sentiment analysis (e.g., Pathao’s Twitter feedback).

B. Data Warehousing

• DSS uses data warehouses to store integrated, historical data for analysis.
• Key features:
  • Subject-Oriented: Organized by topics (e.g., "Customer Behavior," "Sales Trends").
  • Time-Variant: Stores data over time (e.g., monthly sales from 2018–2023).
  • Non-Volatile: Data is never updated or deleted (only added to).
  • Integrated: Combines data from multiple sources (e.g., ERP + external market data).

Worked Example: eSewa’s Fraud Detection DSS

1. Data Sources:
   • Internal: User transaction logs, failed payment records.
   • External: NPR blacklist of fraudulent accounts, global fraud databases.
2. Data Warehouse:
   • Stores 3 years of transaction data in a structured format.
3. Analysis:
   • Data Mining: Detects unusual patterns (e.g., 10 transactions in 1 minute from one IP).
   • Model: Uses anomaly detection algorithms to flag suspicious activity.

4. Model Management in DSS

Models in DSS transform raw data into actionable insights. The two main types are:

A. Mathematical and Statistical Models

Used for quantitative analysis:

• Linear Programming: Optimizes resource allocation (e.g., NTC’s electricity distribution).
• Queuing Theory: Models wait times (e.g., Pathao driver availability).
• Regression Analysis: Predicts trends (e.g., NEPSE stock price forecasting).

Worked Example: Bank Loan Approval DSS Problem: A bank wants to approve loans with minimum risk. Model Used: Logistic Regression (predicts loan default probability).

Visual: Logistic Regression Curve Interpretation: Higher credit scores dramatically reduce default risk.

B. Simulation Models

Used for "what-if" analysis:

• Monte Carlo Simulation: Models uncertainty (e.g., "What if NTC’s tariff increases by 15%?").
• Discrete Event Simulation: Models processes (e.g., Daraz’s order fulfillment delay).

Worked Example: NTC’s Tariff Increase Impact Scenario: NTC proposes a 10% tariff increase. How will it affect customers? Model: Monte Carlo Simulation (runs 1,000 scenarios with random demand fluctuations).

Visual: Simulation Tree (Partial)

5. How DSS Integrates Data and Models

The DSS architecture follows a pipeline from data input to decision output:

flowchart LR
    A["Data Sources\n(Internal/External)"] --> B["Data Warehouse\n(Cleaning, Integration)"]
    B --> C["Model Base\n(Statistical, Simulation, AI)"]
    C --> D["User Interface\n(Dashboards, Reports)"]
    D --> E["Decision Maker\n(Manager, Analyst)"]
    E -->|"Feedback"| A

Key Components:

1. Data Management System (DMS): Handles storage, retrieval, and cleaning.
2. Model Management System (MMS): Stores and executes models.
3. Dialogue System: Allows users to interact with the DSS (e.g., "What if sales tax increases by 5%").

6. Real-World Applications in Nepal

A. eSewa: Fraud Detection (Data-Driven DSS)

• Data Used: Transaction logs, user behavior, NPR blacklists.
• Model: Anomaly detection (identifies unusual spending patterns).
• Impact: Reduces fraud by 30% (2022 report).

B. Ncell: Network Optimization (Model-Driven DSS)

• Data Used: Call drop rates, tower locations, usage patterns.
• Model: Queuing theory to optimize tower placement.
• Impact: Reduces call drops by 25% in Kathmandu.

C. Agricultural Advisory Systems (Knowledge-Driven DSS)

• Data Used: Soil reports, weather data, expert rules.
• Model: Fuzzy logic for pest/disease diagnosis.
• Impact: Increases farmer income by 15% (DoA pilot).

D. NEPSE: Stock Prediction (Hybrid DSS)

• Data Used: Historical prices, market news, economic indicators.
• Models: Time-series forecasting (ARIMA) + NLP for news sentiment.
• Impact: Helps traders make data-backed decisions.

7. Challenges in Data and Model Management

8. Exam Tip: How to Score Full Marks

1. Define Clearly:
   • Always start with definitions (e.g., "Data-driven DSS relies on historical data to support decision-making through OLAP and data mining.").
2. Use Tables for Comparisons:
   • The four DSS types table is highly examinable. Memorize the examples.
3. Show Worked Examples:
   • For model-driven DSS, always include a step-by-step calculation (e.g., logistic regression or simulation steps).
4. Link to Nepal:
   • Every answer must include a Nepalese example (e.g., eSewa, Ncell, NTC).
5. Diagrams > Text:
   • Draw data flow diagrams or model workflows (e.g., the DSS pipeline above).
6. Common Mistakes to Avoid:
   • ❌ Confusing TPS (Transaction Processing System) with DSS.
   • ❌ Forgetting external data sources (e.g., government databases).
   • ❌ Not explaining how models are used (e.g., "OLAP is used for trend analysis").

9. Practice Questions (Self-Assessment)

1. Short Answer:
   • Differentiate between structured and unstructured decisions in DSS. Give one Nepalese example of each.
2. Long Answer:
   • Explain how Ncell uses a model-driven DSS to optimize its network. Include:
     • Data sources.
     • Models applied.
     • Expected outcome.
3. Diagram-Based:
   • Draw the architecture of a web-based DSS used by eSewa for fraud detection. Label all components.

10. Key Formulas to Remember

11. Further Reading

• Books:
  • Decision Support Systems by Efraim Turban (Chapter 5: Data and Model Management).
  • Building Decision Support Systems by Jay Liebowitz.
• Tools:
  • Data Warehousing: SQL Server Analysis Services (SSAS), Oracle OLAP.
  • Modeling: Python (SciKit-Learn), R, Excel Solver.
  • DSS Software: IBM Cognos, SAP BusinessObjects, Tableau.