Little Bits of Artificial Intelligence

1. We’ll simulate a Support Vector Machine (SVM) with 2 features:

• Income (X-axis)
• Credit Card Debt (Y-axis)

The idea is to:

1. Create a few example customers (with labels: good = +1, bad = -1).
2. Manually define a decision boundary (a line).
3. Predict if new applicants are approved or denied loans based on which side of the line they fall on.

Step-by-Step Plan:

1. We’ll define a few points and label them.
2. We’ll assume a decision line: w1 * income + w2 * debt + b = 0
3. Use the sign of the result to predict whether the person is “Good” or “Bad”.

What is Our Classifier Doing?

If the output of w1 * income + w2 * debt + b is:

• > 0 → predict +1 (Good)
• < 0 → predict -1 (Bad)

Python Simulation (Without Libraries)

# === 1. TRAINING DATA ===
# Each customer has: [income, debt]
# Label: +1 = Good (approve loan), -1 = Bad (deny loan)

data = [
    {'income': 60, 'debt': 10, 'label': +1},   # Good customer
    {'income': 80, 'debt': 15, 'label': +1},   # Good customer
    {'income': 30, 'debt': 60, 'label': -1},   # Bad customer
    {'income': 20, 'debt': 80, 'label': -1},   # Bad customer
]

# === 2. HYPOTHETICAL LINE WE CHOSE ===
# Decision boundary: w1 * income + w2 * debt + b = 0
w1 = 1
w2 = -1
b = -10   # Adjust b to move the line up/down

# === 3. PREDICTION FUNCTION ===
def svm_predict(income, debt):
    result = w1 * income + w2 * debt + b
    if result >= 0:
        return +1  # Good customer
    else:
        return -1  # Bad customer

# === 4. TESTING ON TRAINING DATA ===
print("Evaluating known customers:")
for customer in data:
    pred = svm_predict(customer['income'], customer['debt'])
    print(f"Income: {customer['income']}, Debt: {customer['debt']} → Prediction: {'Good' if pred == 1 else 'Bad'}")

# === 5. PREDICTING NEW CUSTOMERS ===
print("\nEvaluating new applicants:")
new_applicants = [
    {'income': 50, 'debt': 20},   # Should be Good
    {'income': 25, 'debt': 70},   # Should be Bad
    {'income': 70, 'debt': 40},   # Close call
]

for applicant in new_applicants:
    pred = svm_predict(applicant['income'], applicant['debt'])
    print(f"Income: {applicant['income']}, Debt: {applicant['debt']} → Prediction: {'Good' if pred == 1 else 'Bad'}")

Output Example

Evaluating known customers:
Income: 60, Debt: 10 → Prediction: Good
Income: 80, Debt: 15 → Prediction: Good
Income: 30, Debt: 60 → Prediction: Bad
Income: 20, Debt: 80 → Prediction: Bad

Evaluating new applicants:
Income: 50, Debt: 20 → Prediction: Good
Income: 25, Debt: 70 → Prediction: Bad
Income: 70, Debt: 40 → Prediction: Good

What This Taught us?

• The decision line was manually crafted — in real SVM, it’s learned using optimization (by maximizing margin).
• The function w1*x + w2*y + b is the core logic of SVM.
• It shows how close the point is to the boundary and on which side it falls.