Classification is a comprehensive perspective that can be adopted in contexts. You may encounter classification methods in machine learning: learn to assign labels of predefined categories to the datasets and incoming data. As an easy-to-understand example, you can consider a task including an animal image classifier into two predefined class labels, the “Cat” and “Dog.”

In Machine Learning, classification methods require two phases named Training and Testing. In addition, we need datasets to implement the Train phase, result evaluation, and final testing. Assume we have a dataset including a couple of hundred images of cats and dogs, and our task is to classify those images into dogs or cats. The training phase is the step in which we teach our model about the differences among classes. We evaluate the result, and the former step might be called for another round until reaching an accurate outcome. The testing step estimates the accuracy of our model on the dataset.

The classification has several types, of which four are the most predominant.

- Binary Classification
- Multi-Class Classification
- Multi-Label Classification
- Imbalanced Classification

Binary Classification

In Binary Classification, there are two class labels, one class label is normal (1), and another class label is abnormal (0). This type is used in the domain similar to YES and NO questions, for example, categorizing a given image into Cat or No. Popular methods that may be referred to as Binary Classification are as below.

- Logistic Regression
- k-Nearest Neighbors
- Decision Trees
- Support Vector Machine
- Naive Bayes

Multi-Class Classification

Multi-Class classification can have two or more class labels, such as the cat and dog labels. The methods mainly used for Binary classification can be adjusted and used in Multi-class classification. They need to pick one of the One-vs-Rest or One-vs-One strategies to be adapted and used in Multi-Class Classification. The popular methods can be counted as below.

- k-Nearest Neighbors
- Decision Trees
- Naive Bayes
- Random Forest
- Gradient Boosting
- Logistic Regression
- Support Vector Machine

Multi-Label Classification

In the Multi-Label classification, we have several class labels with different instances. For example, in our cat and cog datasets, the cat and dog instances are included several breeds in different colors. They should support the attribute of predicting multiple outputs; therefore, the more popular methods include:

- Multi-label Decision Trees
- Multi-label Random Forests
- Multi-label Gradient Boosting

Imbalanced Classification

In this type of classification, the dataset is explicitly unbalanced. Consider the cat and dog dataset with, for example, 500 Cat images and ten dog images. The imbalance causes a disturbance in the prediction process, and the result is fake and phony. In this case, we must pursue additional approaches besides our methods to mitigate the effect of imbalance distribution. Paying more attention to minor classes or Sampling are two main contributions in this area. The famous approaches include as below.

- Random Undersampling.
- SMOTE Oversampling.
- Cost-sensitive Logistic Regression.
- Cost-sensitive Decision Trees.
- Cost-sensitive Support Vector Machines.

To sum up, the class and labels bring about differences in the classification process. We must make sure to deploy the appropriate methods based on the features of the given dataset.