Machine Learning (ML) is a set of techniques that let a computer find patterns in data and use them to make predictions or decisions on new, unseen data — without being explicitly programmed with rules for every case.
The Core Idea
In traditional programming, a human writes explicit rules: if income > 50000 and credit_score > 700: approve_loan(). In machine learning, you flip this around — you give the algorithm historical examples (past loan applications and whether they defaulted), and it learns the rule itself.
| Traditional Programming | Machine Learning | |
|---|---|---|
| Input | Rules + Data | Data + Answers (labels) |
| Output | Answers | Rules (a trained model) |
| Example | Hand-coded tax calculator | Spam filter trained on millions of emails |
This reframing matters because some problems — recognizing a face, predicting churn, detecting fraud — have patterns too complex or too numerous for a human to hand-write as rules, but a model can learn them from examples.
A Minimal Working Example
from sklearn.linear_model import LinearRegression
import numpy as np
# House size (sq ft) -> price (in lakhs INR) — a toy dataset
X = np.array([[500], [750], [1000], [1250], [1500]])
y = np.array([25, 35, 48, 60, 72])
model = LinearRegression()
model.fit(X, y) # learning the pattern from data
predicted = model.predict([[1100]]) # using the pattern on new data
print(predicted)Expected output: approximately [52.9] — the model learned the roughly linear relationship between size and price from 5 examples, then applied it to a size it never saw.
Why Machine Learning, and Why Now
- Data availability: apps, sensors and transactions generate the labeled data ML needs to learn from.
- Compute: GPUs and cloud compute made training on large datasets practical.
- Better algorithms: techniques like gradient boosting and deep learning improved what's learnable from data.
Practical Use Cases
- Email spam filtering, fraud detection, credit scoring
- Product recommendations, search ranking
- Demand forecasting, churn prediction
- Medical image screening, predictive maintenance
Common Mistakes
- Assuming ML is "magic" that works with no data — ML is only as good as the historical data it learns from.
- Confusing machine learning with artificial intelligence in general — see Machine Learning vs AI.
Interview Relevance
Q: "How would you explain machine learning to a non-technical person?" A strong answer uses the traditional-programming-vs-ML contrast above: instead of programming rules, you show the computer examples and it works out the rules itself.
Practice Question
Give two real-world problems: one better solved with traditional hand-coded rules, and one better solved with machine learning. Justify each choice.
