Abstract
The integration of IoT devices in real-time data collection and predictive analytics provides transformative potential toward disaster forecasting and responses. However, existing research faces quite significant challenges including denoising effects, hotspot issues, inefficient resource allocation, overfitting of data, and complexity in decision-making. The proposed solution is a predictive analytic framework for IoT-enabled disaster management, aimed at overcoming these limitations. The methodology is initiated through the construction of a disaster management network by using IoT sensors to generate synthetic data. The collected dataset is denoised using the Denoising Autoencoder (DAE) to get rid of noise and improve data quality. In tuning clustering and resource allocation, we apply the Tuna-Swarm Algorithm (TSA), which aided in efficient management of IoT resources. Switchable Normalization in a custom-designed SN-Convolutional Neutral Network architecture is leveraged for the resources suffering from overfitting. Disaster prediction is based on the Mobile Net-Transformer Generative Model (MN-TGM), an advanced system designed for proper and timely forecasting. The decision-making and resource allocation tasks are subsequently simplified through the use of Fault Tree Analysis as well as Decision-Making Trial and Evaluation Laboratory methods (FTA-DEMATEL). The various performance metrics include latency (ms), overfitting rate (%), prediction accuracy (%), model accuracy (%), and energy efficiency (%) against cluster count. After simulating the system in Ns3.30.1 Ubuntu with Python, its robustness and effectiveness were confirmed. This framework presents a holistic solution to mitigate the existing challenges presented by IoT-enabled disaster management, ensuring timely interventions, exclusive allocation, and improved decision-making, thus averting disaster risks and their impacts.
Index terms: Disaster Management, IoT, prediction, resource allocation, overfitting and decision making.
