Your Gmail decides a message is spam. Swiggy tells you the food will arrive in 34 minutes. Your bank blocks a card the moment someone tries it in another city. Nobody wrote a rule for any of those — the rules were learned from millions of past examples.
That is machine learning, and despite everything happening with generative AI, it is still what most companies actually run and still what most placement interviews ask about. This free course covers the classic ML that comes up in campus and service-company rounds — supervised and unsupervised learning, the algorithms, overfitting, metrics, feature engineering, and how a real project runs from problem to production.
What machine learning actually is
Think about how you would write a spam filter by hand. Block anything containing "free"? You just blocked half your legitimate email. Block anything in capitals? Your aunt writes in capitals. Spammers change tactics every week, so every rule you write is out of date by the time you ship it.
Machine learning is what you do when the rules are too many, too subtle, or too changeable to write. You show the algorithm 100,000 emails already marked spam or not, and it works out the pattern itself.
The flip that defines it
That diagram is the cleanest one-line answer to "what is machine learning?", and interviewers ask it constantly: traditional programming takes rules and data and produces answers; machine learning takes data and answers and produces the rules.
💡 The corollary matters as much as the definition. If a known, exact, stable rule exists — GST on an invoice, whether a password matches — write the rule. An ML model would approximate it, occasionally get it wrong, and be impossible to audit.
Your first model, in ten lines
Predicting house rent from area, using scikit-learn:
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split
X = df[["area_sqft"]] # the feature (input)
y = df["rent"] # the target (what we predict)
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42)
model = LinearRegression()
model.fit(X_train, y_train) # learn from the training data
print("coefficient:", model.coef_[0])
print("test R²: ", model.score(X_test, y_test))
print("predicted rent for 900 sqft:", model.predict([[900]])[0])Result
Four lines of substance. The model learned that each extra square foot adds about ₹42 to the rent — nobody told it that number, it found it.
Notice one thing already: we split the data and scored on the part the model never saw. That habit — measure on held-out data — is the most important discipline in the whole subject, and lesson 5 is about why.
Why learn this for placements
Classic ML questions still dominate technical rounds, even at companies hiring for AI work. Supervised versus unsupervised. What is overfitting. Precision versus recall. Explain the bias-variance tradeoff. Why is accuracy a bad metric for fraud detection. These appear in TCS, Infosys, Wipro, Accenture, Cognizant and Capgemini rounds year after year.
Product companies go further into judgement: which algorithm and why, how would you evaluate it, what would you monitor, and — the question that separates people — where would you not use ML at all?
It is also the foundation under everything else on this site. Generative AI, RAG and fine-tuning all assume you know what training, overfitting and evaluation mean. Pair this with the company exam guides — TCS NQT and Cognizant among them.
How this course works
Ten lessons in learning order. Everyday explanation first, the technical term afterwards, diagrams where the idea is visual, and real numbers rather than hand-waving. No calculus, no derivations — placement interviews ask what a concept means and when to use it, not for a proof.
Every lesson ends with a Practice Zone: six MCQs from real selection rounds with company logos attached, plus two hands-on tasks — diagnosing a model, reading a confusion matrix, finding the leakage in a pipeline. Attempt each one before revealing the solution.
Course roadmap
| # | Lesson | What you walk away with |
|---|---|---|
| 1 | What Is Machine Learning? | the definition, the vocabulary, and when NOT to use ML |
| 2 | Supervised vs Unsupervised | the three learning types and how the data picks one |
| 3 | Regression Explained | linear and logistic regression, R², RMSE vs MAE |
| 4 | Classification Algorithms | trees, forests, boosting, KNN, Naive Bayes, SVM — and which to pick |
| 5 | Overfitting & Underfitting | bias-variance, cross-validation, regularisation, data leakage |
| 6 | Evaluation Metrics | confusion matrix, precision, recall, F1, ROC-AUC — and choosing |
| 7 | Feature Engineering | scaling, encoding, missing values, and the leakage traps |
| 8 | Clustering & PCA | K-Means, choosing k, dimensionality reduction and its cost |
| 9 | Neural Networks Basics | neurons, activations, backpropagation — and when not to use them |
| 10 | ML Project Lifecycle | problem to production, drift, rollout, and the non-technical risks |
Company-wise ML PYQs
After lesson 10 the sidebar continues into company pages — reported questions with full answers: TCS, Infosys, Wipro, Accenture, Cognizant, Capgemini, Amazon and Microsoft.
How to study this course
Read in order — lesson 6 assumes lesson 5, and lesson 10 assumes most of the rest. If you can, keep a notebook open and run the code; the difference between reading "overfitting" and watching your own test score fall while your training score rises is enormous.
If you have one evening: lessons 1, 5 and 6. What ML is, overfitting, and precision versus recall are the three topics that come up in almost every interview, and all three are answerable without writing a line of code.
🎯 The single strongest thing you can bring to an ML interview is one project you can talk about honestly — including what went wrong. "My first version scored 99% and I found a leaked column" is a better answer than a perfect result, because it proves you know how these projects actually fail.
FAQ
Do I need strong mathematics for this?
Not for placement interviews. You need to understand what the concepts mean and when to apply them — this course teaches exactly that and skips the derivations. Mathematics becomes important for research roles and for inventing new methods, not for the questions campus rounds ask.
Is classical ML still relevant now that LLMs exist?
Very much so, and the reason is worth knowing: for structured tabular data — which is most business data — gradient-boosted trees still beat neural networks and LLMs, train in seconds, and are far easier to explain to a regulator. LLMs changed what's possible with text, images and audio. They did not replace the model predicting your electricity demand.
Which programming language and library should I learn?
Python with scikit-learn covers everything in this course and most industry work. Add pandas for data handling, then XGBoost or LightGBM for gradient boosting. PyTorch only when you move into deep learning.
How is this different from the Generative AI course?
This course is about learning patterns from your own data to predict or group things. The Generative AI course is about large pretrained models that produce text. Different problems, different tools — and interviewers frequently ask about both, sometimes in the same round.
Ready? Start here — Lesson 1: What Is Machine Learning? →