Machine Learning (ML)
What is Machine Learning?
Meaning: AI that learns patterns from data, without writing all the rules.
Feature Engineering
Meaning: Humans pick the most important data points ("features") to help the model learn.
Example - Fraud detection:
- Features = transaction amount, time of day, location, device used.
- If someone usually shops in Utah but suddenly makes a $5,000 purchase in London at 3am → suspicious.
Logistic Regression
Meaning: A simple model that predicts probabilities (0-1).
Detailed Example - Bank fraud detection:
- Input: Amount = $2,000, Location = "Paris," Device = "new."
- Output: 0.87 → 87% chance it's fraud.
- Bank flags transaction.
Decision Trees
Meaning: Models shaped like flowcharts with yes/no questions.
Example - Loan approval:
- IF income > $60k AND no late payments → approve.
- ELSE → deny.
- Easy to explain → managers liked them.
Random Forests
Meaning: Many decision trees "vote" together for stronger predictions.
Example: 100 trees predict fraud; 70 say "fraud," 30 say "not fraud." → Final = fraud.
SVM (Support Vector Machine)
Meaning: Draws a "line" (or boundary) to separate two categories of data.
Example: Separate fraudulent vs normal transactions on a graph of amount vs location.
Real-World Examples:
- Fraud Detection: Logistic regression + random forests helped banks stop billions in fraudulent charges.
- Netflix Recommendations: ML used your ratings (features) to suggest movies. If you liked Inception and The Matrix, it suggested other sci-fi thrillers.
Limitation: Works great for tables of numbers, but bad for messy raw data like photos, audio, or text.
The Car Analogy
Classical ML = A car tuned manually by a mechanic.
You need to tell the mechanic exactly what to look for (features), and they adjust settings based on specific rules and patterns they've learned.