Title: Handling Imbalanced Classes With Downsampling
Slug: handling_imbalanced_classes_with_downsampling
Summary: How to handle imbalanced classes with downsampling during machine learning in Python.
Date: 2016-09-06 12:00
Category: Machine Learning
Tags: Preprocessing Structured Data
Authors: Chris Albon

In downsampling, we randomly sample without replacement from the majority class (i.e. the class with more observations) to create a new subset of observation equal in size to the minority class.

Preliminaries



In [1]:

    
# Load libraries
import numpy as np
from sklearn.datasets import load_iris

Load Iris Dataset



In [2]:

    
# Load iris data
iris = load_iris()

# Create feature matrix
X = iris.data

# Create target vector
y = iris.target

Make Iris Dataset Imbalanced



In [3]:

    
# Remove first 40 observations
X = X[40:,:]
y = y[40:]

# Create binary target vector indicating if class 0
y = np.where((y == 0), 0, 1)

# Look at the imbalanced target vector
y









    Out[3]:





array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])

Downsample Majority Class To Match Minority Class



In [4]:

    
# Indicies of each class' observations
i_class0 = np.where(y == 0)[0]
i_class1 = np.where(y == 1)[0]

# Number of observations in each class
n_class0 = len(i_class0)
n_class1 = len(i_class1)

# For every observation of class 0, randomly sample from class 1 without replacement
i_class1_downsampled = np.random.choice(i_class1, size=n_class0, replace=False)

# Join together class 0's target vector with the downsampled class 1's target vector
np.hstack((y[i_class0], y[i_class1_downsampled]))









    Out[4]:





array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])