Machine learning with scikit-learn
Train, test and compare models on tabular data, and explain their results to people who need them.
Created by Lerato Dlamini, Sipho Ndlovu
What you'll learn
- Frame a business question as a prediction problem
- Train regression and classification models with scikit-learn
- Evaluate models with cross-validation, precision and recall
- Build pipelines that stop data leaking between training and testing
- Tune hyperparameters and save a trained model for reuse
- Present model results to a non-technical audience
Skills you'll gain
- scikit-learn
- Model evaluation
- Feature engineering
- pandas
- Cross-validation
- Python
Part of these pathways
Course content
4 sections · 12 lectures · 3.6 total hours · ≈ 1.4 GB to download
How models learn from data
The idea of training, testing and what a model actually learns.
- What supervised learning means Preview 14 min · 105 MB
- Train-test splits and why accuracy can lie Locked until you enrol 20 min · 150 MB
- Fitting a linear regression model Locked until you enrol 22 min · 165 MB
Classification and evaluation
Predict categories and measure how often the predictions are right in the way that matters.
- Logistic regression and decision trees Locked until you enrol 24 min · 180 MB
- Precision, recall and the confusion matrix Locked until you enrol 18 min · 135 MB
- Cross-validation and pipelines Locked until you enrol 22 min · 165 MB
Preparing data and tuning
Get the inputs right and search for better settings.
- Scaling numbers and encoding categories Locked until you enrol 16 min · 120 MB
- Hyperparameter search with GridSearchCV Locked until you enrol 18 min · 135 MB
- Saving and reloading a trained model Locked until you enrol 10 min · < 1 MB
Project: customer churn
A full model from framing to a short presentation.
- Framing the churn problem Locked until you enrol 12 min · 90 MB
- Building and comparing three models Locked until you enrol 25 min · 188 MB
- Presenting results to a manager Locked until you enrol 15 min · < 1 MB
Requirements
- Working Python: loops, functions and pandas basics
- Completed Python for AI and data, or equivalent experience
- Basic statistics such as averages and percentages
Description
This course covers the classic machine learning methods that most business problems still use: regression, classification, decision trees and model evaluation. scikit-learn is the library we use throughout, because it keeps the code short enough to read and explain.
Good results matter less than honest results. You learn why a model can look accurate on training data and fail on new data, how to set up cross-validation and pipelines, and how to report precision, recall and cost in terms that a manager can act on. The final project is a customer churn model built on a realistic dataset.
Instructors
Lerato Dlamini
Senior software engineer and AI trainer
- 9 courses
- 193 learners
Lerato leads the AiX software and AI courses. Lerato has built systems for public-sector clients and teaches the testing and code review habits that the courses depend on, with a focus on code that the next person can maintain.
Sipho Ndlovu
Data analytics lead, SQL and Power BI
- 5 courses
- 110 learners
Sipho runs the AiX data work, from database design to dashboards for provincial reporting. Courses start with the question, then the data, then the chart.
3.9 course rating · 8 ratings
-
Lindiwe Sithole
A good sense of model evaluation: precision, recall, and when accuracy is misleading.
-
Bongani Dube
Solid, but I wish it covered saving and reloading models in more depth.
-
Aisha Hendricks
The pipelines and cross-validation sections are dense. I replayed them three times.
-
Johan Botha
The notebooks broke on a newer scikit-learn release and the fix was not in the course notes.
-
Thabo Mokoena
Intermediate is the right label. You need comfortable Python, and the course says so upfront.
-
Pieter van der Merwe
The churn project used a dataset that looks a lot like what our mobile client handles. Very relevant.
-
Zanele Nkosi
Lerato's tips on feature scaling saved me an afternoon of debugging.
-
Nomsa Khumalo
The train-test split lesson finally explained why my model looked great and then failed on new data.
Frequently asked questions
How much maths do I need?
High-school algebra and an understanding of averages and percentages. Each method is explained with a worked example first.
Which Python version?
Python 3.11 or later with scikit-learn 1.4 or later. The setup lesson lists the exact install commands.
Does the course include a dataset?
Yes. The churn project uses a synthetic telecom dataset that behaves like the real data banks and mobile networks work with.