Free course · Lesson 5 of 5

Real-World ML: Ethics, Data Quality & Deployment

In this lesson you will

Identify sources of bias in data
Explain why human oversight matters

ML is powerful, but it comes with responsibilities. Data quality determines model quality — if your training data is biased, incomplete, or outdated, your model will be too. Ethical ML means ensuring your models are fair, transparent, and accountable.

Data quality: Check for missing values, duplicate records, and outliers. Clean data = better predictions. The old saying applies: "garbage in, garbage out."
Bias in ML: A hiring model trained on historical data where most hires were men will "learn" to prefer men. Always audit your training data for representativeness.
Model interpretability: Can you explain why your model made a certain prediction? Decision Trees are interpretable; deep neural networks are not. For sensitive domains (healthcare, criminal justice), use interpretable models.
Deployment: Save trained models with joblib or pickle. Load them in production to make predictions without retraining. Monitor for "model drift" — when real-world data changes over time.

💡 Practical Skill: Before deploying any ML model, ask: "Who could be harmed if this model makes a mistake?" and "Is my training data representative of the people this model will affect?" Use IBM AI Fairness 360 or Google's What-If Tool to audit your models for bias.

Worked exampleData Quality Check & Saving a Model

Python

# Data quality check + save/load a trained model# Install: pip install scikit-learn joblib import numpy as npfrom sklearn.tree import DecisionTreeClassifierimport joblib  # for saving/loading models # Simulate a dataset with quality issuesdata = [    {"math": 85, "english": 78, "passed": 1},    {"math": 92, "english": 88, "passed": 1},    {"math": None, "english": 65, "passed": 0},  # Missing value!    {"math": 45, "english": 50, "passed": 0},    {"math": 70, "english": None, "passed": 1},  # Missing value!    {"math": 95, "english": 92, "passed": 1},    {"math": 30, "english": 40, "passed": 0},    {"math": 88, "english": 85, "passed": 1},] print("=== Data Quality Check ===")missing_math = sum(1 for d in data if d["math"] is None)missing_english = sum(1 for d in data if d["english"] is None)print(f"Missing math scores: {missing_math}")print(f"Missing english scores: {missing_english}") # Clean data: drop rows with missing values (simple approach)clean_data = [d for d in data if d["math"] is not None and d["english"] is not None]print(f"Rows after cleaning: {len(clean_data)} (removed {len(data) - len(clean_data)})") # Prepare features and labelsX = np.array([[d["math"], d["english"]] for d in clean_data])y = np.array([d["passed"] for d in clean_data]) # Train a modelmodel = DecisionTreeClassifier(max_depth=3)model.fit(X, y) # Save the model to a filejoblib.dump(model, "student_pass_model.pkl")print("\n✅ Model saved as 'student_pass_model.pkl'") # Load the model (in a real deployment, this would be a separate script)loaded_model = joblib.load("student_pass_model.pkl") # Predict a new studentnew_student = np.array([[75, 80]])prediction = loaded_model.predict(new_student)[0]print(f"New student (Math: 75, English: 80) => {'PASS' if prediction == 1 else 'FAIL'}") # Feature importanceprint(f"Math importance: {model.feature_importances_[0]:.3f}")print(f"English importance: {model.feature_importances_[1]:.3f}")

Practice exercise

Describe how a model trained only on urban school data could fail rural students, and one check you would run before using it.

Key takeaways

Models learn the biases in their data
Check performance separately for important groups
Use models to support people's decisions, not replace them

In the full program Module 5: Responsible ML and a First Project See how this topic is taught in depth, with practical activities and an assessment.

Free self-study lesson — no certificate. For the full instructor-led program, see registration options.