Skip to content

Projects: Module 01 — Introduction to AI and Machine Learning

Projects that apply the concepts from this module. These are starter projects — build them for real, not just as exercises.

Project 1: End-to-End Iris Classification Pipeline (Beginner)

Time estimate: 3–4 hours Skills practiced: Full ML workflow, EDA, train/test split, evaluation

Build a clean, reusable Python script (not a notebook) that:

  1. Loads the Iris dataset
  2. Performs EDA: prints shape, class distribution, and per-feature statistics
  3. Splits with stratification
  4. Scales features correctly (no leakage)
  5. Trains a DecisionTreeClassifier
  6. Produces a classification_report and saves it to a text file
  7. Takes max_depth as a command-line argument so it is easy to experiment

Acceptance criteria: - Running python iris_classifier.py --max_depth 3 produces a clean report - The preprocessing is correctly implemented (no data leakage) - The script handles the case where max_depth is too large (document the expected behavior)

Project 2: Bias-Variance Explorer (Intermediate)

Time estimate: 4–6 hours Skills practiced: Bias-variance tradeoff, validation curves, model comparison

Build a Python script that empirically demonstrates the bias-variance tradeoff on a dataset of your choice.

Requirements: 1. Use a real dataset (not Iris — try the wine, breast cancer, or digits datasets from sklearn) 2. Train a DecisionTreeClassifier for each depth from 1 to 25 3. For each depth, compute and store: training accuracy, 5-fold cross-validation accuracy 4. Print a formatted table showing both metrics for every depth 5. Identify and print: the depth with best CV accuracy, the depth where overfitting begins 6. Write a short (3–5 paragraph) analysis of your results

Bonus: Repeat the same experiment with a KNeighborsClassifier varying n_neighbors. How does the bias-variance tradeoff manifest differently for a nearest-neighbor model vs. a tree?

Project 3: ML Problem Classification Guide (Beginner)

Time estimate: 2–3 hours Skills practiced: Problem framing, paradigm selection, metric selection

Create a Markdown reference document that classifies 10 real-world ML applications. For each application:

  1. State the application in one sentence (e.g., "Predict whether an email is spam")
  2. Identify the ML paradigm (supervised/unsupervised/reinforcement) and justify it
  3. Identify the specific problem type (classification/regression/clustering/etc.)
  4. State what the input features would be
  5. State what the label/target is (or what structure you're looking for)
  6. Identify the primary evaluation metric and why

Choose applications from at least 3 different domains (healthcare, finance, media, e-commerce, science, etc.).

This project develops the problem-framing skill that practitioners use before writing any code.