Anomaly Detection with PyCaret

Anomaly Detection with PyCaret, Unsupervised learning: Anomaly Detection with PyCaret Workflow.

Anomaly detection identifies outliers in any given situation. Used for a wide range of use cases – to identify fraud in financial services, and for predictive maintenance in manufacturing, for identifying fake news in social media management, understanding the intuition behind anomaly detection is a critical tool in every data scientist’s toolbox.

The course begins with an introduction to Anomaly Detection:

1. The types of Anomalies
2. Anomaly detection use cases
3. The intuition behind some of the anomaly detection algorithms: Isolation Forest, Local Outlier Factor, and KNN

In the second part of the course, we go through a discussion on the PyCaret workflow:

1. How the PyCaret library simplifies data-cleaning and preparation for anomaly detection
2. The range of anomaly detection algorithms available
3. How to assign models
4. How to visualize the results of anomaly detection in PyCaret.

In the third and final part of the course, we work with an inbuilt PyCaret social media dataset (the ‘Facebook’ dataset):

1. We first undertake exploratory data analysis using Python Seaborn
2. We identify anomalies based on the reactions to posts/videos/links and other content types etc. In this case, the problem statement is to identify content that might need to be reviewed owing to the disproportionate number of reactions.
3. We work with a handful of anomaly detection models, and examine the dataset for the observations which are flagged as anomalous.
4. We discover that these are content types that have received a large number of reactions, and the content types and reaction types vary from algorithm to algorithm.