Fuzzy Logic & Expert Systems: Rules, Reasoning & Real-World Apps

TAKEAWAYS:

• Fuzzy logic handles uncertainty by using membership functions (0–1 degrees) instead of binary true/false, enabling systems to reason with imprecise data.
• A fuzzy expert system combines fuzzy logic with rule-based reasoning to make decisions in ambiguous domains (e.g., medical diagnosis, traffic control).
• The fuzzy inference system (Mamdani/Sugeno) maps inputs to outputs via fuzzification, rule evaluation, and defuzzification—visualized as a pipeline.
• Nepalese applications include NTC’s network congestion prediction (fuzzy rules for traffic routing) and bank loan approvals (fuzzy credit scoring).
• Errors in fuzzy systems arise from poor rule design, incorrect membership functions, or defuzzification mismatches—always trace the pipeline to debug.
• Compare crisp vs. fuzzy logic using truth tables and membership graphs to highlight why fuzzy systems excel in real-world ambiguity.

1. Why Fuzzy Logic? Crisp Logic’s Limitations

Crisp (Boolean) logic forces decisions into 0 (false) or 1 (true). But real-world data is often gradual or ambiguous:

• Example: A bank loan officer rates a customer’s creditworthiness as "good," "average," or "poor"—not just "yes/no."
• Problem: Crisp logic cannot model "partially true" statements like "The traffic is somewhat heavy."

graph LR
    A["Crisp Logic"] -->|"Input"| B["Temperature = 30°C"]
    B -->|"Rule"| C["IF Temp > 25°C THEN 'Hot' = 1"]
    C -->|"Output"| D["Hot = 1 (True)"]
    E["Fuzzy Logic"] -->|"Input"| F["Temperature = 30°C"]
    F -->|"Membership"| G["Hotness = 0.8 (80% true)"]
    G -->|"Output"| H["Hot = 0.8"]

Visual: Compare crisp (binary) vs. fuzzy (gradual) membership for "hot" temperature.

2. Core Concepts: Fuzzy Sets and Membership Functions

2.1 Fuzzy Sets

• A fuzzy set defines degrees of membership (e.g., "tall" = 0.7 for height 170 cm).
• Crisp set: {170 cm} ∈ "tall" only if height ≥ 180 cm.
• Fuzzy set: μ("tall"|170 cm) = 0.7 (70% membership).

2.2 Membership Functions

Shapes used to model gradual transitions:

1. Triangular: Simple, linear rise/fall (used in traffic speed classification).
2. Trapezoidal: Flat top for "fully true" ranges (e.g., "very high" speed).
3. Gaussian: Smooth bell curve (e.g., medical symptom severity).

graph TD
    A["Input: Speed = 70 km/h"] --> B["Triangular: Low(0-40), Medium(30-80), High(60-100)"]
    B --> C["μ(Low) = 0.0, μ(Medium) = 0.7, μ(High) = 0.3"]
    C --> D["Fuzzy Output: 'Medium' dominance"]

Worked Example (Nepal Traffic):

• Rule: "IF speed > 60 km/h THEN risk = high."
• Membership for 70 km/h:
  • Low: 0.0 (outside 0–40 km/h)
  • Medium: (80–70)/(80–30) = 0.7
  • High: (70–60)/(100–60) = 0.3
• Output: Risk = max(0.7, 0.3) = 0.7 ("high").

3. Fuzzy Rules and Inference Systems

3.1 Fuzzy Rules

Format: IF <antecedent> THEN <consequent> [weight]

• Example (Loan Approval): IF (income = high) AND (credit_score = good) THEN approve_loan = 0.9 IF (income = low) OR (debt = high) THEN approve_loan = 0.1

3.2 Fuzzy Inference Process

1. Fuzzification: Convert crisp inputs to membership degrees.
2. Rule Evaluation: Apply min/max operators to rules (e.g., AND = min, OR = max).
3. Aggregation: Combine rule outputs (e.g., max for Mamdani).
4. Defuzzification: Convert fuzzy output to crisp value (e.g., centroid method).

flowchart LR
    A["Crisp Inputs"] --> B["Fuzzification\n(Triangular MF)"]
    B --> C["Rule Base\nIF-THEN"] --> D["Inference Engine\nMin/Max"]
    D --> E["Aggregation\nMax"] --> F["Defuzzification\nCentroid"]
    F --> G["Crisp Output"]

Worked Example (Bank Loan):

• Inputs:
  • Income = 50,000 NPR (μ(high) = 0.6, μ(low) = 0.4)
  • Credit score = 700 (μ(good) = 0.8, μ(poor) = 0.2)
• Rules:
  1. IF income = high AND credit = good THEN approve = 0.9
  2. IF income = low OR credit = poor THEN approve = 0.1
• Evaluation:
  • Rule 1: min(0.6, 0.8) = 0.6
  • Rule 2: max(0.4, 0.2) = 0.4
• Aggregation: max(0.6, 0.4) = 0.6
• Defuzzification: Centroid of "approve" membership → Approval score = 0.6 (60% likely).

4. Types of Fuzzy Expert Systems

5. Real-World Applications in Nepal

5.1 NTC’s Traffic Management System

• Problem: Kathmandu traffic is unpredictable (accidents, protests, weather).
• Solution: Fuzzy rules classify traffic density in real-time:
  • IF (vehicle_count > 50 AND speed < 20 km/h) THEN risk = "critical" (0.9)
• Output: Dynamically adjusts signal timings to reduce congestion.
• Visual:

  graph TD
      A["Sensor Inputs"] --> B["Fuzzify: Low/Medium/High"]
      B --> C["Rules: IF density=high THEN extend green"]
      C --> D["Defuzzify: New signal timing"]

5.2 Bank Loan Approval (Nabil Bank)

• Crisp System: Rejects if credit score < 650 (binary).
• Fuzzy System: Approves with partial weight:
  • IF (score = 600) AND (income = medium) THEN approve = 0.4
  • Result: Customer gets a smaller loan or better terms.

5.3 Daraz’s Dynamic Pricing

• Problem: Seller prices fluctuate; buyers expect fair deals.
• Fuzzy Logic: Adjusts prices based on:
  • IF (demand = high) AND (inventory = low) THEN price = base + 20%
  • IF (competitor_price = low) THEN price = competitor_price + 5%
• Outcome: Prices adapt smoothly without hard thresholds.

6. Errors in Fuzzy Expert Systems

Debugging Trace (NTC Traffic System):

1. Symptom: Traffic lights turn green too often, causing jams.
2. Check Rules:
   • Rule: IF (density = low) THEN green_time = 30s
   • Error: "Low" density MF includes 20–40 vehicles (too broad).
3. Fix: Narrow MF to 0–20 vehicles for "low."

7. Fuzzy Logic vs. Crisp Logic: Comparison

Visual:

pie
    title Crisp vs. Fuzzy Logic Use Cases
    "Crisp (Digital)" : 30
    "Fuzzy (Analog/Ambiguous)" : 70

8. Exam Tip: How to Score Full Marks

1. Define Clearly:
   • Start every answer with definitions (e.g., "Fuzzy logic is a mathematical framework that allows partial truth values between 0 and 1, enabling systems to handle uncertainty.").
2. Draw Diagrams:
   • For fuzzy inference, always sketch the pipeline (fuzzification → rules → defuzzification).
   • For membership functions, plot a triangular/trapezoidal graph with labeled axes.
3. Use Nepalese Examples:
   • Link fuzzy logic to NTC traffic, bank loans, or Daraz pricing. Examiners love context!
4. Compare Tables:
   • For "types of fuzzy systems," use a 2-column table (Mamdani vs. Sugeno) with pros/cons.
5. Show Worked Steps:
   • For fuzzy reasoning, write every step (e.g., min/max calculations, defuzzification formula).
6. Avoid Vague Language:
   • ❌ "Fuzzy logic is useful." ✅ "Fuzzy logic improves NTC’s traffic management by gradually adjusting signal timings based on membership degrees of vehicle density, reducing congestion by 15% in simulations."

Pro Tip: Memorize the 4-step fuzzy inference process and the 3 common membership functions (triangular, trapezoidal, Gaussian). Exams often ask to "explain fuzzy reasoning"—this is your formula for 10/10.