Efficient resource management within Internet of Things (IoT) environments remains a pressing challenge due to the increasing number of devices and their diverse functionalities. This study introduces a neural network-based model that uses Long-Short-Term Memory (LSTM) to optimize resource allocation under dynamically changing conditions. Designed to monitor the workload on individual IoT nodes, the model incorporates long-term data dependencies, enabling adaptive resource distribution in real time. The training process utilizes Min-Max normalization and grid search for hyperparameter tuning, ensuring high resource utilization and consistent performance. The simulation results demonstrate the effectiveness of the proposed method, outperforming the state-of-the-art approaches, including Dynamic and Efficient Enhanced Load-Balancing (DEELB), Optimized Scheduling and Collaborative Active Resource-management (OSCAR), Convolutional Neural Network with Monarch Butterfly Optimization (CNN-MBO), and Autonomic Workload Prediction and Resource Allocation for Fog (AWPR-FOG). For example, in scenarios with low system utilization, the model achieved a resource utilization efficiency of 95% while maintaining a latency of just 15 ms, significantly exceeding the performance of comparative methods.