Techniques to find outliers

Data Science, AI and ML
#1

What are the techniques to find outliers?

#2

One from box plot, Z-score

#3

Unfortunately, there are no hard and fast rules that everyone agrees on.

A relatively simple method is visually inspecting the plot and discerning a general pattern for the points (maybe it’s elliptical and oriented in some direction) and tossing out (with note) points that don’t fit the pattern. If this seems somewhat subjective, it is.

If a data point diverges from the general trend of the graph it can be termed as an outlier and an inspection for the assignable causes can be carried out at that point. However if the sample is expected to be distributed randomly, outliers are detected by fixing the acceptable limits and the points that lie beyond these limits are then classified as outliers. The most generally used limits are 3-sigma limits. In this, the acceptable values must lie between 3 standard deviation below and above the expected value. Any point lying beyond 3-sigma is termed as an outlier and is checked for the assignable causes.