Machine Learning with Imbalanced Data

Name: Machine Learning with Imbalanced Data
Rating: 4.7 (858 reviews)

Learn to over-sample and under-sample your data, apply SMOTE, ensemble methods, and cost-sensitive learning.

Created bySoledad Galli, Train in Data Team

Last updated 9/2024

English

What you'll learn

Apply random under-sampling to remove observations from majority classes
Perform under-sampling by removing observations that are hard to classify
Carry out under-sampling by retaining observations at the boundary of class separation
Apply random over-sampling to augment the minority class
Create syntethic data to increase the examples of the minority class
Implement SMOTE and its variants to synthetically generate data
Use ensemble methods with sampling techniques to improve model performance
Change the miss-classification cost optimized by the models to accomodate minority classes
Determine model performance with the most suitable metrics for imbalanced datasets

Course content

10 sections • 106 lectures • 9h 15m total length

Course Curriculum Overview3:11
Course Material1:42
Use Jupyter notebooks to implement techniques for imbalanced datasets, complemented by videos, presentations, and online datasets. Read three articles to download code from GitHub, presentations, and datasets before you begin.
Code | Jupyter notebooks0:33
Presentations covered in the course0:19
Python package Imbalanced-learn0:08
Download Datasets0:08
Resources to learn machine learning skills0:40

Introduction to Performance Metrics3:22
Accuracy4:21
Accuracy - Demo5:39
Compare accuracy of random forest and logistic regression against a baseline majority-predictor using scikit-learn, emphasizing minority-class detection and concluding logistic regression best for positives.
Precision, Recall and F-measure13:32
Install Yellowbrick0:07
Precision, Recall and F-measure - Demo10:04
Compare precision, recall, and F1 scores for imbalanced data using random forest and logistic regression, with scikit-learn and yellowbrick visuals, and optimize thresholds to improve minority-class performance.
Confusion tables, FPR and FNR6:03
Confusion tables, FPR and FNR - Demo7:32
Balanced Accuracy3:49
Explore how balanced accuracy addresses imbalanced datasets by averaging recall across classes. Examine confusion matrix insights, including true positives, and recall to compare class performance beyond overall accuracy.
Balanced accuracy - Demo2:43
Geometric Mean, Dominance, Index of Imbalanced Accuracy4:29
Geometric Mean, Dominance, Index of Imbalanced Accuracy - Demo9:28
Apply geometric mean, dominance, and the index of imbalanced accuracy to compare random forest and logistic regression on imbalanced data, using recall, true negative rate, and balanced accuracy.
ROC-AUC7:26
ROC-AUC - Demo4:46
Precision-Recall Curve11:19
Precision-Recall Curve - Demo3:53
Additional reading resources (Optional)0:16
Probability4:32

Under-Sampling Methods - Introduction5:21
Random Under-Sampling - Intro4:23
Random Under-Sampling - Demo10:11
Explore how random under-sampling balances imbalanced data with imbalanced-learn, preserves original distributions, and boosts model performance using a random forest on toy and real datasets.
Condensed Nearest Neighbours - Intro8:03
Condensed Nearest Neighbours - Demo7:25
Tomek Links - Intro4:43
Explore Tomek links, pairs of nearest-neighbor observations from opposite classes. Remove the majority observation or the entire TomekLinks pair to reduce boundary noise and improve learning.
Tomek Links - Demo3:05
One Sided Selection - Intro4:38
One Sided Selection - Demo3:00
Edited Nearest Neighbours - Intro5:01
Discover edited nearest neighbors, a cleaning method that undersamples the majority class by removing samples near the boundary using a 3-nearest neighbors check.
Edited Nearest Neighbours - Demo4:02
Repeated Edited Nearest Neighbours - Intro4:39
Repeated Edited Nearest Neighbours - Demo3:00
All KNN - Intro6:16
All KNN - Demo5:50
Neighbourhood Cleaning Rule - Intro6:14
Neighbourhood Cleaning Rule - Demo1:55
Apply the neighbourhood cleaning rule with fit_resample to undersample the majority class, using a 3-neighbour rule and a two-to-one threshold.
NearMiss - Intro3:47
NearMiss - Demo3:53
Instance Hardness - Intro9:20
Explore instance hardness threshold and filtering to remove hard observations, reduce class overlap, and use probability thresholds or percentiles with one-vs-rest for multiclass classification.
Instance Hardness Threshold - Demo16:21
Instance Hardness Threshold Multiclass Demo7:44
Undersampling Method Comparison7:44
Wrapping up the section5:18
Setting up a classifier with under-sampling and cross-validation10:54
Train the random forest on the under-sampled training set and evaluate on the test set with the original class distribution using cross-validation and the average precision-recall score.
Summary Table0:28

Over-Sampling Methods - Introduction3:41
Random Over-Sampling5:00
Explore random over-sampling to balance imbalanced data by duplicating minority-class observations to a 1:1 ratio, using imbalanced-learn's RandomOverSampler and fit_resample, and adjust balancing strategy for multiclass targets.
Random Over-Sampling - Demo4:55
ROS with smoothing - Intro6:39
Explore random over-sampling with smoothing to generate new minority-class examples by adding noise guided by class distribution, controlled by a shrinkage factor, and implemented with Imbalanced-learn's RandomOverSampler.
ROS with smoothing - Demo4:36
Explore random over-sampling with smoothing using imbalanced-learn's RandomOverSampler to create synthetic minority-class examples and observe dispersion changes as the shrinkage factor varies.
SMOTE9:26
SMOTE - Demo2:35
SMOTE-NC9:02
SMOTE-NC - Demo2:56
SMOTE-N19:25
SMOTE-N Demo7:20
ADASYN7:11
ADASYN - Demo3:17
Borderline SMOTE7:47
borderline smote refines smote by generating synthetic minority samples near the decision boundary, using knn to select the danger group and offering two variants for interpolation and extrapolation.
Borderline SMOTE - Demo3:13
SVM SMOTE16:40
Resources on SVMs0:07
SVM SMOTE - Demo4:32
K-Means SMOTE13:01
K-Means SMOTE - Demo3:29
Over-Sampling Method Comparison5:50
Wrapping up the section9:30
compare over-sampling versus under-sampling for imbalanced data, discuss SMOTE and variants (ADASYN, border-line SMOTE, SVM SMOTE, K-means SMOTE), and emphasize distance metrics and practical trade-offs.
How to Correctly Set Up a Classifier with Over-sampling5:24
Setting Up a Classifier - Demo4:13
Implement oversampling schemes with cross-validation using Imbalanced-learn make_pipeline and scikit-learn cross_validate; compare SMOTE and other oversamplers with a random forest and MinMaxScaler, highlighting varied dataset results.
Summary Table0:06

Combining Over and Under-sampling - Intro6:02
Combining Over and Under-sampling - Demo5:26
Comparison of Over and Under-sampling Methods5:54
Combine over and under-sampling manually0:05
Wrapping up2:08
Conclude that there is no universal rule for over-sampling or under-sampling; test original data, random over-sampling, or combinations, while noting nearest-neighbour under-sampling scalability and distance metrics for categorical features.

Ensemble methods with Imbalanced Data4:49
Explore ensemble methods for imbalanced data by combining bagging and boosting with data level techniques such as under- and over-sampling, including hybrid ensembles and cost-sensitive strategies, with Python implementations.
Foundations of Ensemble Learning3:12
Bagging3:04
Bagging, or bootstrap aggregating, creates datasets by sampling with replacement and trains classifiers on each. It combines predictions by averaging or voting to improve generalization and foster diverse, de-correlated models.
Bagging plus Over- or Under-Sampling5:38
Boosting10:03
Boosting plus Re-Sampling7:05
Explore boosting with data pre-processing to handle imbalanced data, including RUSBoost, SMOTEBoost, RAMOBoost, and ADASYN-inspired RAMOBoost, focusing on under-sampling, synthetic samples, and weighted ensembles.
Hybdrid Methods4:48
Explore hybrid methods that combine bagging, boosting, and resampling to boost model performance on imbalanced data, including Balanced Random Forests, EasyEnsemble, and BalanceCascade using AdaBoost.
Ensemble Methods - Demo9:59
Compare ensemble methods with and without resampling on imbalanced data, using random forests, boosting, bagging, and easy ensemble in scikit-learn and imbalanced-learn, and evaluate ROC-AUC across datasets.
Wrapping up5:31
Additional Reading Resources0:17

Cost-sensitive Learning - Intro7:27
Types of Cost10:55
Obtaining the Cost4:28
Cost Sensitive Approaches1:52
Misclassification Cost in Logistic Regression3:35
Misclassification Cost in Decision Trees4:02
Cost Sensitive Learning with Scikit-learn7:13
Discover cost-sensitive learning in scikit-learn by applying class_weight or sample_weight to misclassification costs. See how balancing techniques and custom class penalties improve ROC-AUC.
Find Optimal Cost with hyperparameter tuning3:33
Learn to treat the misclassification cost as a hyperparameter and optimize it with grid search on a random forest using the KDD 2004 dataset, balancing ratio, and ROC-AUC evaluation.
Additional Reading Resources0:13

Requirements

Knowledge of machine learning basic algorithms, i.e., regression, decision trees and nearest neighbours
Python programming, including familiarity with NumPy, Pandas and Scikit-learn
A Python and Jupyter notebook installation

Description

Welcome to Machine Learning with Imbalanced Datasets. In this course, you will learn multiple techniques which you can use with imbalanced datasets to improve the performance of your machine learning models.

If you are working with imbalanced datasets right now and want to improve the performance of your models, or you simply want to learn more about how to tackle data imbalance, this course will show you how.

We'll take you step-by-step through engaging video tutorials and teach you everything you need to know about working with imbalanced datasets. Throughout this comprehensive course, we cover almost every available methodology to work with imbalanced datasets, discussing their logic, their implementation in Python, their advantages and shortcomings, and the considerations to have when using the technique. Specifically, you will learn:

Under-sampling methods at random or focused on highlighting certain sample populations
Over-sampling methods at random and those which create new examples based of existing observations
Ensemble methods that leverage the power of multiple weak learners in conjunction with sampling techniques to boost model performance
Cost sensitive methods which penalize wrong decisions more severely for minority classes
The appropriate metrics to evaluate model performance on imbalanced datasets

By the end of the course, you will be able to decide which technique is suitable for your dataset, and / or apply and compare the improvement in performance returned by the different methods on multiple datasets.

This comprehensive machine learning course includes over 50 lectures spanning more than 10 hours of video, and ALL topics include hands-on Python code examples which you can use for reference and for practice, and re-use in your own projects.

In addition, the code is updated regularly to keep up with new trends and new Python library releases.

So what are you waiting for? Enroll today, learn how to work with imbalanced datasets and build better machine learning models.

Who this course is for:

Data scientists and machine learning engineers working with imbalanced datasets
Data scientists who want to improve the performance of models trained on imbalanced datasets
Students who want to learn intermediate content on machine learning
Students working with imbalanced multi-class targets

Machine Learning with Imbalanced Data

What you'll learn

Explore related topics

Course content

Introduction7 lectures • 7min