Unsupervised Learning Using K-means algorithm

Abstract

This research aims to apply the k-means algorithm, which is considered one of the most important unsupervised learning techniques in clustering unlabeled data. We tested its application on a two-dimensional dataset consisting of (300,2) taken from the Kaggle platform. We downloaded the data and then manually specified the number of clusters K=3, as specifying the clusters is the main problem in the algorithm. We also specified the number of iterations T=6, and the results showed that the algorithm gradually improved across iterations. We used evaluation metrics to assess the performance of the algorithm, where we used the objective function, which decreased from 4719.65 to 266.65 at final stability. The Cohesion metrics also showed a significant decrease, reflecting that the points are interconnected within each cluster. The Separation evaluation metric shows the distance between clusters. These results indicate the effectiveness of the algorithm in dividing data into clusters in a short time and with high efficiency. However, relying on manually entering the number of clusters K is a major problem for the algorithm and requires further solutions. Therefore, future work should explore methods to solve this problem, especially in the case of large data sets, such as using the Elbow method, as combining such methods enhances the results and selects k in a non-manual way, making the clustering process more accurate and effective. aggregation process more accurate and effective.