After data gathering, they are normally grouped in clusters with similar objects depending on the desired data application. With the rapid development in technology across all fields, data mining has become an essential tool. For example, it has helped in the running of various economic sectors by making well-informed decisions and enabling prediction of future market trends. Other than data mining research, data clustering is also an important aspect of statistical data analysis. Up to date, discovering interesting groups in a given data set has remained the greatest challenge in data clustering.
Among the available data clustering techniques, the K-means algorithm is widely used. Being sensitive to the initial points, this technique is limited to spherical clusters and may converge to the local minima since its performance depends on the initial cluster centers. Considering the above challenges, researchers have been looking for alternative solutions and have identified the use of meta-heuristic algorithms as a potential solution. This approach has not only overcome the shortcoming of traditional algorithms but also provided new ideas for solving clustering problems.
Recently, Shilei Qiao, Yongquan Zhou, Professor Yuxiang Zhou, and Dr. Rui Wang from the Guangxi University of Nationalities particularly assessed the feasibility of the simple water cycle algorithm with percolation for solving clustering problems. This is a type of meta-heuristic algorithms that is based on the real-world observation of the water cycle process including the flow of streams and rivers into the sea. The main objective was to look for optimal solutions and overcome most of the challenges associated with the original water-cycle algorithm such as low accuracy and slow convergence speed. The algorithm was based on discarding the process of rainfall to simplify its process and evolutionary process. The work is currently published in the research journal, Soft Computing.
The evolutionary process was controlled by the percolation operator and the flowing process. The algorithm’s ability to fully explore the search space more accurately was demonstrated by the process of streamflow to the river and the river flow to the sea. In addition to searching and finding the best solution accurately, the percolation operator generated new streams thus increasing the population diversity. This may be used to represent the local search while the process of flowing can be used on behalf of the global search to increase the convergence speed.
To prove the concept, the performance of the proposed algorithm was evaluated using selected ten data sets and the results compared with other data clustering techniques. The authors observed that PWCA has the potential to converge to the global optimum value at a relatively smaller standard deviation and higher precision. This resulted in very clear and accurate clustering effects.
In general, the proposed algorithm performed better than its counterparts in terms of accuracy, speed, quality, stability, and reliability of the final solutions. This makes it a more robust technique for data clustering and analysis. Furthermore, the study will enable future elaboration and testing of numerous real clustering problems and a large number of patterns.
Qiao, S., Zhou, Y., Zhou, Y., & Wang, R. (2019). A simple water cycle algorithm with percolation operator for clustering analysis. Soft Computing, 23(12), 4081-4095.Go To Soft Computing