Imbalanced Data & Anomaly Detection

About the course

We often face the problem of looking for a needle in a haystack of data. A tedious and challenging task that requires specific tools to tackle. Enter Imbalanced Data and Anomaly Detection, a training badge to equip you with two specific tools to do so. One is aimed at working with imbalanced datasets in supervised learning, for instance by creating many more needles so they are easier to find. The other will help you to identify what a needle is in the first place, using unsupervised anomaly detection methods. The first day will begin laying the foundation of anomaly detection by exploring the context and different types of anomalies, discussing those unique to your business and testing your skills in detecting anomalies using data plotting. We will then look at the first challenge: imbalanced datasets. Learning why they are considered an issue and how to maximize model results from imbalanced datasets. The second day will delve into anomaly detection. Beginning with unsupervised learning, looking at its application and evaluating the effectiveness. Then moving onto time-series anomaly detection, using Z-score, decomposition, and forecasting.  Of course, this training would not be complete without ensuring your newly found knowledge can be translated into reality. Thus, the final hurdle will be an analytical task whereby you apply what you have learned to detect anomalies in a real-life case.  

Why this is for you

You might be confident at building models and interpreting the results, however, without knowledge of these evaluation metrics, you most likely will obtain naïve results rendering your efforts useless. Equipping yourself with these tools for detecting anomalies and imbalanced data will empower you to be confident in your models and applying them in business.  

For whom

This training course is designed with both Data Scientists and Data Engineers in mind. At Expert level, this training requires previous knowledge and the completion of our Classification Using Tree Models course (3202).  

What you’ll learn

This training will develop your skills and draw on a variety of tools and methods for the detection and transformation of anomalies and imbalanced data. Specific content covered includes:

1. Point, contextual, and collective anomalies
2. Univariate and multivariate anomalies
3. Structured and unstructured anomaly detection
4. Graphical tools such as jitter plots, violin plots, Z-scores and more
5. Cost-sensitive learning
6. Static and dynamic anomaly detection
7. Z-score, modified Z-score, decomposition and forecasting for anomaly detection
8. Two unsupervised algorithms: Isolation Forest, One-Class SVM

Learning Goals

• Introduction to anomaly detection: Able to explain the context of anomalies and distinguish between different types.
• Graphical anomaly detection: Able to detect anomalies using data plotting.
• Handling imbalanced datasets: Able to maximize model results from imbalanced datasets.
• Unsupervised anomaly detection: Knowing when unsupervised learning can be applied and how to evaluate an unsupervised model.
• Time-series anomaly detection: Able to detect anomalies in time-series data.
• Unsupervised algorithms: Able to apply two unsupervised algorithms to detect anomalies.
• Detecting fraud using all anomaly detection tools: Able to combine all your knowledge to detect anomalies in a real-life case.

  Theory and practical use Each training in the GAIn portfolio combines high-quality standardized training material with theory sessions from experts and hands-on experience where you directly apply the material to real-life cases. Each training is developed by top of the field practitioners which means they are full of industry examples along with practical challenges and know-how, fueling the interactive discussions during training. We believe this multi-level approach creates the ideal learning environment for participants to thrive.