Machine Learning Roadmap
Goal
- Understand core ML concepts deeply
- Be able to build and evaluate models from scratch
- Finish with real mini-projects
RULES (read once, follow always)
-
Don’t watch videos passively
-
Every topic → code it yourself
-
Always:
Learn → Code → Break → Fix → Extend -
If something feels “easy”, you didn’t go deep enough
PHASE 1 — CLASSIFICATION (Week 1)
Main Resource
Topics + Reinforcement
1. Binary Classification
-
Understand:
- Features vs Labels
- What is a class
📺 Watch:
💻 Do:
-
Create dataset manually
-
Predict:
- pass/fail
- spam/not spam
2. Supervised Learning Workflow
-
MUST understand pipeline:
load data → split → fit → predict → evaluate
💻 Do:
- Implement full workflow from scratch (no copy-paste)
3. k-Nearest Neighbors (KNN)
Core idea:
- Distance-based learning
💻 Do:
- Change K values (1, 3, 5, 10)
- Observe accuracy changes
📺 Reinforce:
4. Model Performance
-
Learn:
- Accuracy
- Why accuracy can lie
💻 Do:
- Create imbalanced dataset
- Test accuracy
5. Train/Test Split
💻 Do:
-
Try:
- 50/50
- 80/20
- 90/10
👉 Observe changes
6. Overfitting vs Underfitting
💻 Do:
- Small dataset → overfit
- Large dataset → better generalization
7. Model Complexity
💻 Do:
- Increase K → underfitting
- Decrease K → overfitting
Mini Project (MANDATORY)
Pick ONE:
- Student pass prediction
- Spam classifier
- Movie genre classifier
PHASE 2 — REGRESSION (Week 2)
Main Resource
Topics
1. Linear Regression
💻 Do:
-
Predict:
- house prices
- student scores
2. Model Evaluation
-
Learn:
- MSE
- R²
3. Cross Validation
💻 Do:
-
Compare:
- with CV
- without CV
4. Regularization
- Ridge
- Lasso
💻 Do:
- Apply both and compare results
Mini Project
-
Predict:
- expenses
- sales
- grades
PHASE 3 — MODEL EVALUATION & TUNING (Week 3)
📘 Main Resource
Topics
1. Metrics
- Precision
- Recall
- F1-score
2. Logistic Regression
💻 Do:
- Binary classification again with logistic regression
3. ROC Curve & AUC
- Understand model quality
4. Hyperparameter Tuning
💻 Do:
- GridSearchCV
- RandomizedSearchCV
Mini Project
- Improve your Phase 1 classifier using tuning
PHASE 4 — PREPROCESSING & PIPELINES (Week 4)
Main Resource
- DataCamp Chapter 4
Topics
1. Data Cleaning
- Missing values
- Encoding
2. Scaling
- Standardization
3. Pipelines
💻 Do:
-
Build full pipeline:
preprocess → model → evaluate
Final Project (IMPORTANT)
Build ONE solid project:
- Student performance predictor
- Budget predictor
- Song popularity predictor
👉 Include:
- preprocessing
- model
- evaluation
- tuning
EXTRA RESOURCES (USE SMARTLY)
1. Documentation (VERY IMPORTANT)
Use it when:
- You don’t understand a function
- You want real-world usage
2. YouTube Strategy
Use your playlists like this:
- Stanford → concepts only
- Second playlist → implementation
Don’t binge
3. Practice (Non-negotiable)
- Use small datasets
- Try modifying everything
WHAT YOU SHOULD MASTER BY THE END
If you can do these without help, you’re good:
- Build ML pipeline from scratch
- Choose correct model
- Evaluate properly
- Avoid overfitting
- Tune hyperparameters