ML | Handling Imbalanced Data with SMOTE and Near Miss Algorithm in Python

In Machine Learning and Data Science we often come across a term called Imbalanced Data Distribution, generally happens when observations in one of the class are much higher or lower than the 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.

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. In more technical words, if we have imbalanced data distribution in our dataset then our model becomes more prone to the case when minority class has negligible or very lesser recall.

Imbalanced Data Handling Techniques: There are mainly 2 mainly algorithms that are widely used for handling imbalanced class distribution.

1. SMOTE
2. Near Miss Algorithm

SMOTE (Synthetic Minority Oversampling Technique) – Oversampling

SMOTE (synthetic minority oversampling technique) is one of the most commonly used oversampling methods to solve the imbalance problem.
It aims to balance class distribution by randomly increasing minority class examples by replicating them.
SMOTE synthesises new minority instances between existing minority instances. It generates the virtual training records by linear interpolation for the minority class. These synthetic training records are generated by randomly selecting one or more of the k-nearest neighbors for each example in the minority class. After the oversampling process, the data is reconstructed and several classification models can be applied for the processed data.
More Deep Insights of how SMOTE Algorithm work !

Step 1: Setting the minority class set A, for each , the k-nearest neighbors of x are obtained by calculating the Euclidean distance between x and every other sample in set A.

Step 2: The sampling rate N is set according to the imbalanced proportion. For each , N examples (i.e x1, x2, …xn) are randomly selected from its k-nearest neighbors, and they construct the set .

Step 3: For each example (k=1, 2, 3…N), the following formula is used to generate a new example:

in which rand(0, 1) represents the random number between 0 and 1.

NearMiss Algorithm – Undersampling

NearMiss is an under-sampling technique. It aims to balance class distribution by randomly eliminating majority class examples. 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.
To prevent problem of information loss in most under-sampling techniques, near-neighbor methods are widely used.
The basic intuition about the working of near-neighbor methods is as follows:

For finding n closest instances in the majority class, there are several variations of applying NearMiss Algorithm :

1. NearMiss – Version 1 : It selects samples of the majority class for which average distances to the k closest instances of the minority class is smallest.
2. NearMiss – Version 2 : It selects samples of the majority class for which average distances to the k farthest instances of the minority class is smallest.
3. NearMiss – Version 3 : It works in 2 steps. Firstly, for each minority class instance, their M nearest-neighbors will be stored. Then finally, the majority class instances are selected for which the average distance to the N nearest-neighbors is the largest.

This article helps in better understanding and hands-on practice on how to choose best between different imbalanced data handling techniques.

Load libraries and data file

The dataset consists of transactions made by credit cards. This dataset has 492 fraud transactions out of 284, 807 transactions. That makes it highly unbalanced, the positive class (frauds) account for 0.172% of all transactions.
The dataset can be downloaded from here.

Output:

RangeIndex: 284807 entries, 0 to 284806
Data columns (total 31 columns):
Time      284807 non-null float64
V1        284807 non-null float64
V2        284807 non-null float64
V3        284807 non-null float64
V4        284807 non-null float64
V5        284807 non-null float64
V6        284807 non-null float64
V7        284807 non-null float64
V8        284807 non-null float64
V9        284807 non-null float64
V10       284807 non-null float64
V11       284807 non-null float64
V12       284807 non-null float64
V13       284807 non-null float64
V14       284807 non-null float64
V15       284807 non-null float64
V16       284807 non-null float64
V17       284807 non-null float64
V18       284807 non-null float64
V19       284807 non-null float64
V20       284807 non-null float64
V21       284807 non-null float64
V22       284807 non-null float64
V23       284807 non-null float64
V24       284807 non-null float64
V25       284807 non-null float64
V26       284807 non-null float64
V27       284807 non-null float64
V28       284807 non-null float64
Amount    284807 non-null float64
Class     284807 non-null int64

Output:

       0    284315
       1       492

Split the data into test and train sets

Output:

      Number transactions X_train dataset:  (199364, 29)
      Number transactions y_train dataset:  (199364, 1)
      Number transactions X_test dataset:  (85443, 29)
      Number transactions y_test dataset:  (85443, 1)

Now train the model without handling the imbalanced class distribution

Output:

                precision   recall   f1-score  support

           0       1.00      1.00      1.00     85296
           1       0.88      0.62      0.73       147

    accuracy                           1.00     85443
   macro avg       0.94      0.81      0.86     85443
weighted avg       1.00      1.00      1.00     85443

The accuracy comes out to be 100% but did you notice something strange ?
The recall of the minority class in very less. It proves that the model is more biased towards majority class. So, it proves that this is not the best model.
Now, we will apply different imbalanced data handling techniques and see their accuracy and recall results.

Using SMOTE Algorithm

You can check all the parameters from here.

Output:

Before OverSampling, counts of label '1': [345]
Before OverSampling, counts of label '0': [199019] 

After OverSampling, the shape of train_X: (398038, 29)
After OverSampling, the shape of train_y: (398038, ) 

After OverSampling, counts of label '1': 199019
After OverSampling, counts of label '0': 199019

Look! that SMOTE Algorithm has oversampled the minority instances and made it equal to majority class. Both categories have equal amount of records. More specifically, the minority class has been increased to the total number of majority class.
Now see the accuracy and recall results after applying SMOTE algorithm (Oversampling).

Prediction and Recall

Output:

                precision   recall   f1-score  support

           0       1.00      0.98      0.99     85296
           1       0.06      0.92      0.11       147

    accuracy                           0.98     85443
   macro avg       0.53      0.95      0.55     85443
weighted avg       1.00      0.98      0.99     85443

Wow, We have reduced the accuracy to 98% as compared to previous model but the recall value of minority class has also improved to 92 %. This is a good model compared to the previous one. Recall is great.
Now, we will apply NearMiss technique to Under-sample the majority class and see its accuracy and recall results.

NearMiss Algorithm:

You can check all the parameters from here.

Output:

Before Undersampling, counts of label '1': [345]
Before Undersampling, counts of label '0': [199019] 

After Undersampling, the shape of train_X: (690, 29)
After Undersampling, the shape of train_y: (690, ) 

After Undersampling, counts of label '1': 345
After Undersampling, counts of label '0': 345

The NearMiss Algorithm has undersampled the majority instances and made it equal to majority class. Here, the majority class has been reduced to the total number of minority class, so that both classes will have equal number of records.

Prediction and Recall

Output:

               precision    recall   f1-score   support

           0       1.00      0.56      0.72     85296
           1       0.00      0.95      0.01       147

    accuracy                           0.56     85443
   macro avg       0.50      0.75      0.36     85443
weighted avg       1.00      0.56      0.72     85443

This model is better than the first model because it classifies better and also the recall value of minority class is 95 %. But due to undersampling of majority class, its recall has decreased to 56 %. So in this case, SMOTE is giving me a great accuracy and recall, I’ll go ahead and use that model! 🙂