Comparison of Unsupervised Learning Algorithms for K-Means Clustering Efficiency in IoE Sensor Detection Systems for Smart Agriculture

This study investigates the application of clustering algorithms to data collected from color sensors providing frequency signals during the cultivation process. Continuous sampling allowed for real-time data collection, and variations in temperature and humidity were observed to influence signal reception. The dataset's large volume and varying sample sizes necessitated the evaluation of several clustering models for efficient data processing. The experiment tested K-Means, K-Means++, Mini-Batch K-Means, and Bisecting K-Means algorithms. The results reveal that the computation time for each model increased with the number of groups. Specifically, the K-Means, K-Means++, and Mini-Batch K-Means models exhibited shorter computation times, ranging from 0.02 to 0.03 seconds for two-group scenarios, while the Bisecting K-Means algorithm took significantly longer, approximately 0.10 seconds. Despite the variation in processing times, all models performed similarly in terms of clustering quality, with an average Silhouette Score of 0.82. This score indicates that the algorithms effectively separated data points into well-defined groups while maintaining high accuracy in clustering. Additionally, the analysis highlighted the importance of balancing computation time and clustering quality, particularly as the number of clusters increases. These findings suggest that while computational efficiency may vary across models, the clustering quality remains robust, providing valuable insights for data analysis in agricultural sensor networks. The study's results demonstrate that each clustering model is capable of handling large datasets with high accuracy in grouping, making them applicable to real-world agricultural data processing scenarios. In conclusion, the experiment confirms that all tested models, despite differing in computation time, successfully partitioned the data with high-quality results, offering promising applications for data analysis and optimization in agriculture and sensor-based systems.