Kishor.Academy
Log in
← All courses

Artificial Intelligence · intermediate

Applied Machine Learning: Build, Evaluate, Improve

Fit a real regression model, compare a baseline, calculate meaningful metrics, investigate errors, and write an honest model card.

With Admin

intermediate9 lessons · 3 sections8–12 hours of reading and experiments, plus 3–5 hours for the project; estimates exclude video.

THE TAKEAWAYS

What you’ll learn

  • Define target, features, prediction cutoff, and a relevant success measure.
  • Fit a regression line and compare a training-only baseline on the same held-out cases.
  • Explain feature engineering and fit preprocessing only within training data.
  • Calculate MAE and interpret accuracy and recall under class imbalance.
  • Use validation and error analysis without repeatedly tuning on final test results.
  • Identify subgroup evaluation, drift monitoring, and fallback needs.
  • Compare controlled experiments and record data, settings, and run outputs.
  • Write a model card with intended use, limitations, evaluation, and decision ownership.
  • Submit a reproducible predictor, baseline comparison, error analysis, and model card.

Is this for you?

Learners with basic Python and data-cleaning skills who want to build and evaluate their first predictive model.

Before you start

  • Complete Python & Data or be comfortable with functions, CSV validation, and train/test splits.
  • Python 3.10 or newer; basic algebra, averages, and percentages.
  • Use the supplied synthetic data and document actual run outputs. No GPU, cloud account, or paid API is required.

PUT IT TO WORK

What you’ll build

Workshop attendance predictor. Client scenario: estimate attendance from registrations for materials planning. Deliverable: executable model, baseline comparison, residual table, model card, and experiment record.

Define a prediction problem before selecting a model. Fit a least-squares line to fictional workshop records, compare it with a training-mean baseline on the same held-out rows, and inspect residuals. Explore why accuracy can mislead for imbalanced classification and why changing the evaluation split can invalidate a comparison. Produce a model card that clearly separates a classroom result from evidence needed for real deployment.

Course curriculum

SECTION 1Prediction problems, baselines, and model fitting3 lessons
  1. Frame the problem before choosing a modelReading
  2. Regression and classification baselinesReading
  3. Feature engineering and pipelinesReading
SECTION 2Metrics, errors, and reliable comparisons3 lessons
  1. Accuracy is not enoughReading
  2. Cross-validation and error analysisReading
  3. Fairness, drift, and monitoringReading
SECTION 3Iteration, deployment limits, and a model card3 lessons
  1. Experiment tracking and reproducibilityReading
  2. Model cards and decision logsReading
  3. Capstone: end-to-end prediction projectReading