Integrating Disaster Data Clustering with Neural Networks for Comprehensive Analysis

Abstract

The rise in both natural and human-induced disasters has caused considerable harm to the inhabitants of our planet. These disasters, ranging from tsunamis and floods to earthquakes and forest fires, not only wreak havoc on property but also tragically claim human lives. With the increasing accessibility of large datasets and the growing prominence of parallel computing architectures, clustering algorithms are once again taking center stage. Spectral Clustering, in particular, has proven to outperform many traditional clustering methods. It transforms the clustering problem into a graph-partitioning challenge by treating each data point as a graph node. What sets Spectral Clustering apart is its ability to handle a broader range of clustering problems without imposing specific data assumptions. Despite its ease of implementation and computational efficiency, it can be time-consuming for dense datasets due to matrix construction and eigenvalue calculations. BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies), an unsupervised data mining algorithm, is designed for hierarchical clustering in large datasets. A notable advantage of BIRCH is its adaptive and incremental clustering approach for multi-dimensional metric data points, aiming for optimal clustering quality within available resources. In the domain of disaster data analysis, neural networks play a pivotal role. An implemented neural network architecture for the forest fire dataset demonstrated impressive accuracy (97%) with minimal loss. This underscores the significant contribution of neural networks in addressing challenges related to disaster data analysis