Analysis of Clustering Algorithms with Explainable AI and Optimized Attributes of Flood Dataset

Clustering is one of the most popular unsupervised learning techniques in machine learning. It is used to recognize hidden patterns in data and group them accordingly. Data can be categorized using different clustering algorithms. However, there is no specific clustering algorithm that can be certified as the best in all cases. It is a matter of study which algorithm suits the best for a particular dataset. However, in contrast to the current literature, no study has been found to perform a cluster analysis based on a flood dataset, which has a significant impact on flood risk management. In this study, we have compared 10 clustering algorithms: k-means, bisecting k-means, fuzzy c-means, Gaussian mixture model (GMM), agglomerative, BIRCH, spectral, DBSCAN, OPTICS and mean shift according to 5 evaluation metrics: Rand index, silhouette score, Fowlkes-Mallows index, Davies-Bouldin index and Calinski-Harabasz index by applying them to a flood dataset in order to categorize data to identify flood. Moreover, optimal features from the flood dataset were selected based on random forest algorithm. The k-means is the best method for the dataset because it showed the best performance according to three of the five assessment measures. Furthermore, we extend our analysis to include the SHAP value analysis along with the means to interpret the results.