Remove outliers using numpy. Normally, an outlier is outside 1.5 * the IQR experimental analysis has shown that a higher/lower IQR might produce more accurate results. Interestingly, after 1000 runs, removing outliers creates a larger standard deviation between test run results.

outlier_removal.py

import numpy as np

def removeOutliers(x, outlierConstant):
    a = np.array(x)
    upper_quartile = np.percentile(a, 75)
    lower_quartile = np.percentile(a, 25)
    IQR = (upper_quartile - lower_quartile) * outlierConstant
    quartileSet = (lower_quartile - IQR, upper_quartile + IQR)
    resultList = []
    for y in a.tolist():
        if y >= quartileSet[0] and y <= quartileSet[1]:
            resultList.append(y)
    return resultList
    

Thanks, @adrian-alberto! Updated

How can this piece of code be adopted for a dataframe? to drop values across the dataframe

Thanks for posting, I need this code for dataframe too. I will try to modify it for my case .

How do I decide what the constant is ?

Hi, here is my suggestion to take advantage of numpy's speed instead of a python loop with a growing list. With big arrays the difference in time is noticeable.

def removeOutliers(x, outlierConstant):
    a = np.array(x)
    upper_quartile = np.percentile(a, 75)
    lower_quartile = np.percentile(a, 25)
    IQR = (upper_quartile - lower_quartile) * outlierConstant
    quartileSet = (lower_quartile - IQR, upper_quartile + IQR)
    
    result = a[np.where((a >= quartileSet[0]) & (a <= quartileSet[1]))]
    
    return result.tolist()

Thanks, @adrian-alberto! Updated

Did you mean 0.25 and 0.75 rather than 25 and 75? Percentiles go from 0 to 100. Thanks for the code.

@marcoruizrueda
What you are talking about are quantiles.

0 quartile = 0 quantile = 0 percentile
1 quartile = 0.25 quantile = 25 percentile
2 quartile = .5 quantile = 50 percentile (median)
3 quartile = .75 quantile = 75 percentile
4 quartile = 1 quantile = 100 percentile

what is outlier constant?