Introduction to Resampling methods

While reading about Machine Learning and Data Science we often come across a term called Imbalanced Class Distribution , generally happens when observations in one of the classes are much higher or lower than any other classes. 
As Machine Learning algorithms tend to increase accuracy by reducing the error, they do not consider the class distribution. This problem is prevalent in examples such as Fraud Detection, Anomaly Detection, Facial recognition etc.
Two common methods of Resampling are – 

1. Cross Validation
2. Bootstrapping

Cross Validation –

Cross-Validation is used to estimate the test error associated with a model to evaluate its performance.
Validation set approach: 
This is the most basic approach. It simply involves randomly dividing the dataset into two parts: first a training set and second a validation set or hold-out set. The model is fit on the training set and the fitted model is used to make predictions on the validation set. 

Leave-one-out-cross-validation: 
LOOCV is a better option than the validation set approach. Instead of splitting the entire dataset into two halves only one observation is used for validation and the rest is used to fit the model. 

k-fold cross-validation –
This approach involves randomly dividing the set of observations into k folds of nearly equal size. The first fold is treated as a validation set and the model is fit on the remaining folds. The procedure is then repeated k times, where a different group each time is treated as the validation set. 

Bootstrapping –

Bootstrap is a powerful statistical tool used to quantify the uncertainty of a given model. However, the real power of bootstrap is that it could get applied to a wide range of models where the variability is hard to obtain or not output automatically.
Challenges: 
Algorithms in Machine Learning tend to produce unsatisfactory classifiers when handled with unbalanced datasets. 
For example, Movie Review datasets

Total Observations : 100
Positive Dataset : 90
Negative Dataset : 10
Event rate : 2%

The main problem here is how to get a balanced dataset.
Challenges with standard ML algorithms: 
Standard ML techniques such as Decision Tree and Logistic Regression have a bias towards the majority class, and they tend to ignore the minority class. They tend only to predict the majority class, hence, having major misclassification of the minority class in comparison with the majority class.
Evaluation of classification algorithm is measured by confusion matrix. 

A way to evaluate the results is by the confusion matrix, which shows the correct and incorrect predictions for each class. In the first row, the first column indicates how many classes “True” got predicted correctly, and the second column, how many classes “True” were predicted as “False”. In the second row, we note that all class “False” entries were predicted as class “True”. 
Therefore, the higher the diagonal values of the confusion matrix, the better the correct prediction. 
Handling Approach:

• Random Over-sampling: 
  It aims to balance class distribution by randomly increasing minority class examples by replicating them.
  For example –

Total Observations : 100
Positive Dataset : 90
Negative Dataset : 10
Event Rate : 2%

• We replicate Negative Dataset 15 times 

Positive Dataset: 90
Negative Dataset after Replicating: 150
Total Observations: 190
Event Rate : 150/240= 63%

• SMOTE (Synthetic Minority Oversampling Technique) synthesises new minority instances between existing minority instances. It randomly picks up the minority class and calculates the K-nearest neighbour for that particular point. Finally, the synthetic points are added between the neighbours and the chosen spot.
• Random Under-Sampling: 
  It aims to balance class distribution by randomly eliminating majority class examples.
  For Example – 

Total Observations : 100
Positive Dataset : 90
Negative Dataset : 10
Event rate : 2%

We take 10% samples of Positive Dataset and combine it with Negative Dataset.

Positive Dataset after Random Under-Sampling : 10% of 90 = 9 

Total observation after combining it with Negative Dataset: 10+9=19

Event Rate after Under-Sampling : 10/19 = 53%

• When instances of two different classes are very close to each other, we remove the instances of the majority class to increase the spaces between the two classes. This helps in the classification process.
• Cluster-based Over Sampling: 
  K means clustering algorithm is independently applied to both the class instances such as to identify clusters in the datasets. All clusters are oversampled such that clusters of the same class have the same size.
  For Example – 

Total Observations : 100
Positive Dataset : 90
Negative Dataset : 10
Event Rate : 2%

• Majority Class Cluster: 
  Cluster 1: 20 Observations 
  Cluster 2: 30 Observations 
  Cluster 3: 12 Observations 
  Cluster 4: 18 Observations 
  Cluster 5: 10 Observations
  Minority Class Cluster: 
  Cluster 1: 8 Observations 
  Cluster 2: 12 Observations
  After oversampling all clusters of the same class have the same number of observations.
  Majority Class Cluster: 
  Cluster 1: 20 Observations 
  Cluster 2: 20 Observations 
  Cluster 3: 20 Observations 
  Cluster 4: 20 Observations 
  Cluster 5: 20 Observations
  Minority Class Cluster: 
  Cluster 1: 15 Observations 
  Cluster 2: 15 Observations 
  
  Below is the implementation of some resampling techniques: 
  You can download the dataset from the given link below : Dataset download

Python3

# importing libraries
import pandas as pd
import numpy as np
import seaborn as sns
from sklearn.preprocessing import StandardScaler
from imblearn.under_sampling import RandomUnderSampler, TomekLinks
from imblearn.over_sampling import RandomOverSampler, SMOTE

Python3

dataset = pd.read_csv(r'C:\Users\Abhishek\Desktop\creditcard.csv')



print("The Number of Samples in the dataset: ", len(dataset))
print('Class 0        :', round(dataset['Class'].value_counts()[0]
                      /len(dataset) * 100, 2), '% of the dataset')

print('Class 1(Fraud) :', round(dataset['Class'].value_counts()[1]
                      /len(dataset) * 100, 2), '% of the dataset')

Python3

X_data = dataset.iloc[:, :-1]
Y_data = dataset.iloc[:, -1:]

rus = RandomUnderSampler(random_state = 42)
X_res, y_res = rus.fit_resample(X_data, Y_data)

X_res = pd.DataFrame(X_res)
Y_res = pd.DataFrame(y_res)


print("After Under Sampling Of Major Class Total Samples are :", len(Y_res))
print('Class 0        :', round(Y_res[0].value_counts()[0]
                /len(Y_res) * 100, 2), '% of the dataset')

print('Class 1(Fraud) :', round(Y_res[0].value_counts()[1]
                /len(Y_res) * 100, 2), '% of the dataset')

Python3

tl = TomekLinks()

X_res, y_res = tl.fit_resample(X_data, Y_data)

X_res = pd.DataFrame(X_res)
Y_res = pd.DataFrame(y_res)


print("After TomekLinks Under Sampling Of Major Class Total Samples are :", len(Y_res))
print('Class 0        :', round(Y_res[0].value_counts()[0]
                /len(Y_res) * 100, 2), '% of the dataset')

print('Class 1(Fraud) :', round(Y_res[0].value_counts()[1]
                /len(Y_res) * 100, 2), '% of the dataset')

Python3

ros = RandomOverSampler(random_state = 42)

X_res, y_res = ros.fit_resample(X_data, Y_data)

X_res = pd.DataFrame(X_res)
Y_res = pd.DataFrame(y_res)


print("After Over Sampling Of Minor Class Total Samples are :", len(Y_res))
print('Class 0        :', round(Y_res[0].value_counts()[0]
                /len(Y_res) * 100, 2), '% of the dataset')

print('Class 1(Fraud) :', round(Y_res[0].value_counts()[1]
                /len(Y_res) * 100, 2), '% of the dataset')

Python3

sm = SMOTE(random_state = 42)

X_res, y_res = sm.fit_resample(X_data, Y_data)

X_res = pd.DataFrame(X_res)
Y_res = pd.DataFrame(y_res)


print("After SMOTE Over Sampling Of Minor Class Total Samples are :", len(Y_res))
print('Class 0        :', round(Y_res[0].value_counts()[0]
                 /len(Y_res) * 100, 2), '% of the dataset')

print('Class 1(Fraud) :', round(Y_res[0].value_counts()[1]
                 /len(Y_res) * 100, 2), '% of the dataset')