Median Absolute Deviation

The MAD (Median Absolute Deviation) rule is a statistical outlier detection method that uses the median and the median absolute deviation as robust estimators to identify outliers in a dataset.

It is particularly useful when dealing with data that may not follow a normal distribution or when there are potential outliers that can significantly impact the mean and standard deviation.

How to use MAD rule

Calculate the Median: Compute the median of the dataset, which is the middle value when the data is sorted;
Calculate the Median Absolute Deviation (MAD): For each data point, find the absolute difference between the data point and the median. The MAD is the median of these absolute differences;
Define a Threshold: Choose a threshold value (usually a constant, e.g., 2 or 3 times the MAD) to determine how far a data point can deviate from the median before being considered an outlier;
Identify Outliers: Any data point that has an absolute difference from the median greater than the threshold is considered an outlier.

Note

Mathematically, the absolute difference between two values, A and B, is denoted as |A - B|, where "|" represents the absolute value function. This function returns the positive value of the difference between A and B.

MAD rule implementation


python

MAD vs 1.5 IQR rule

Var allt tydligt?

Tack för dina kommentarer!

Avsnitt 2. Kapitel 5

Fråga AI

Fråga vad du vill eller prova någon av de föreslagna frågorna för att starta vårt samtal

Kursinnehåll

Data Anomaly Detection

1. What is Anomaly Detection?

General Information Types of Anomalies How Outliers Influence On Prediction Results What Should We Do With Detected Outliers

2. Statistical Methods in Anomaly Detection

Rule-based Approach Challenge: Rule-based Approach 1.5 IQR Rule 3-Sigma Rule Median Absolute Deviation Challenge: Outlier Detection Using MAD Rule

3. Machine Learning Techniques

Clustering Challenge: Using DBSCAN Clustering to Detect Outliers Regularisation Challenge: Solving Task Using Regularisation Autoencoders