Classification Analysis

Let’s explore binary classification analysis, which categorizes data into two types. Classification analysis is also a type of supervised learning. Here, we’ll examine simple two-dimensional (X, Y) data. On the graph, multiple Xs are located at the top, and multiple Os are at the bottom. The goal of binary classification is to use known X and O data to determine the category of unknown data.

First, we need to identify a linear function that separates the boundary between X and O. The objective is to find the values of W and b that satisfy Y=W×X+b. The technique is similar to the concept of linear regression discussed earlier. Finding a linear function that separates the categories is not particularly challenging.

Classification Analysis

The question is how to apply this linear function to binary classification. This is where the activation function comes in. A common activation function is the sigmoid function. The sigmoid function is a simple concept that converts input data into values between 0 and 1 based on a certain threshold. It’s widely used in machine learning because it transforms linear data into non-linear data. Here, if the known value y is greater than the calculated y^ from the linear function, it is converted to 1; otherwise, it is converted to 0.

A Quick Note

• Difference Between Linear and Non-Linear

Linear data implies that it is close to a straight-line distribution, making it relatively easy to predict. Non-linear data, on the other hand, does not follow a straight line and may converge to extreme values or take a curved shape. Non-linear data is more difficult to predict than linear data. Most data used in machine learning is non-linear. Therefore, the initial calculation uses a linear function, and then an activation function is applied to add non-linearity for a more detailed representation of the data's characteristics.

When first encountering machine learning, one may feel overwhelmed by various mathematical expressions. However, for Developers, understanding the basic principles and usage of machine learning is more relevant. Basic knowledge of arithmetic operations, logarithms, and matrix operations is sufficient to grasp fundamental concepts. Rather than focusing on the equations, it may be helpful to observe the shape of the graphs that these equations represent.

Binary Classification Function

A binary classification function can be mathematically represented as shown above. Besides the sigmoid function, other activation functions such as ReLU and tanh are also used. The concepts introduced in classification analysis form the foundation for understanding neural networks, which will be covered next. By approaching machine learning and deep learning concepts gradually, starting with the basics, anyone can master them without feeling overwhelmed.