Free to read ยท Three levels

Machine Learning: trustworthy evaluation and model workflows

A complete theory course from task framing and leakage-free preprocessing to models, uncertainty, fairness, explanations, deployment, and drift.

18 lessonsEnglish + ArabicContent reviewed
Start learning

In this course

Learn machine learning as an explicit prediction and decision process rather than an algorithm leaderboard. Eighteen substantial lessons connect data meaning, valid splits, baselines, model families, selection, calibration, subgroup evidence, and operational ownership. Each lesson includes explained examples, diagnostic mistakes, and complete worked exercises.

All executable examples use small fixed synthetic inputs and local Python/NumPy calculations. Their printed metrics demonstrate arithmetic under the stated fixture only; they are not claimed external accuracy, research replications, or deployment guarantees. No external dataset, downloaded model, GPU, cloud resource, or live service is required.

What you will learn

  • Frame a prediction task and choose a legitimate baseline and split.
  • Fit preprocessing and selection without leaking evaluation information.
  • Explain regression, classification, trees, ensembles, clustering, and reduction objectives.
  • Assess uncertainty, calibration, subgroup harms, and explanation limits.
  • Package validated inference contracts and investigate drift under controlled retraining.

Your learning path

Three levels

Prediction and data foundations

Tasks, data quality, splits, preprocessing, regression, and classification.

  1. Problem framing and useful baselines
  2. Data quality and feature contracts
  3. Splits, groups, time, and leakage
  4. Preprocessing without leakage
  5. Regression, loss, and linear models
  6. Classification scores and metrics
Explore this level

Models and controlled selection

Optimization, trees, ensembles, clustering, dimensionality reduction, and cross-validation.

  1. Optimization and regularization
  2. Trees and partitioning
  3. Ensembles and error diversity
  4. Clustering and distance
  5. Dimensionality reduction and selection
  6. Cross-validation and tuning
Explore this level

Evidence and operational responsibility

Uncertainty, calibration, fairness, explanations, deployment contracts, and drift.

  1. Evaluation uncertainty and model comparisons
  2. Calibration and decision costs
  3. Fairness, slices, and potential harms
  4. Explanations and causal limits
  5. Deployment contracts and reproducibility
  6. Monitoring, drift, and retraining
Explore this level

Who this is for

  • Python learners with data-analysis foundations.
  • Developers and analysts building reliable predictive evaluation and inference workflows.

Before you start

  • Python functions, arrays, exceptions, and basic NumPy indexing.
  • Descriptive statistics, means, variance, conditional rates, and basic linear algebra.
  • Data-analysis concepts including joins, missingness, units, and reproducible scripts.

Related technical solutions

Investigate suspicious validation results and leakageTrace what information each fitted step and selection decision could see before trusting a surprisingly strong score.Resolve inference schema and preprocessing mismatchesCompare the same synthetic record through the stored training-side transform and inference boundary before blaming the estimator.Investigate strong offline scores and weak operational outcomesSeparate pipeline mismatch, evaluation design, population change, delayed labels, and decision policy before deciding whether retraining is justified.