Cloud computing environments demand efficient and dynamic resource allocation strategies to meet fluctuating user demands and optimize resource utilization. This paper proposes an Adaptive Neuro-Fuzzy Inference System (ANFIS) integrated with Reinforcement Learning (RL), specifically Q-learning, for enhanced dynamic resource allocation in cloud environments. The proposed system, ANFIS-RL, leverages the learning capabilities of RL to adapt the fuzzy rules of ANFIS, enabling it to dynamically adjust resource allocation policies based on real-time system performance and changing workload patterns. We detail the architecture of ANFIS-RL, the Q-learning algorithm implementation, and the experimental setup used to evaluate its performance. The results demonstrate that ANFIS-RL significantly outperforms traditional rule-based fuzzy systems and standard Q-learning approaches in terms of resource utilization, response time, and overall system efficiency. This hybrid approach offers a robust and adaptive solution for managing the complexities of resource allocation in dynamic cloud environments.
Time series forecasting is a critical task across various domains, including finance, weather prediction, and supply chain management. However, the inherent non-linearity and non-stationarity of many real-world time series pose significant challenges for traditional forecasting models. This paper proposes an enhanced Adaptive Neuro-Fuzzy Inference System (ANFIS) optimized with metaheuristic algorithms for improved time series forecasting accuracy. Specifically, we investigate the integration of Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) to optimize the parameters of the ANFIS architecture. The proposed hybrid approaches, ANFIS-PSO and ANFIS-GA, are compared against standard ANFIS and other benchmark time series forecasting methods using several real-world datasets. Experimental results demonstrate that the metaheuristic-optimized ANFIS models achieve significantly superior forecasting accuracy, particularly in handling complex and volatile time series data. The integration of metaheuristics enhances the adaptability and robustness of ANFIS, leading to more reliable and accurate predictions.
Social media platforms have become crucial sources of information for understanding public opinion. However, the inherent noisiness and contextual complexity of social media data pose significant challenges for accurate sentiment analysis. This paper presents a novel hybrid attention-based deep learning model designed to address these challenges. Our approach combines the strengths of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), augmented with multiple attention mechanisms, to effectively capture both local and global contextual information. The model also incorporates a pre-processing module for noise reduction and normalization. Extensive experiments on benchmark datasets demonstrate that our proposed model outperforms state-of-the-art sentiment analysis techniques, particularly in handling noisy and contextually ambiguous social media text. The results highlight the effectiveness of the hybrid architecture and attention mechanisms in capturing nuanced sentiment expressions, contributing to more accurate and robust sentiment analysis systems.
The Internet of Things (IoT) is rapidly expanding, connecting billions of devices and transforming various aspects of our lives. However, this interconnectedness introduces significant security challenges, making IoT networks vulnerable to various cyberattacks. Traditional security measures are often inadequate to address the complex and evolving threat landscape. This paper proposes a hybrid deep learning framework for enhanced intrusion detection in IoT networks. The framework integrates feature engineering techniques to extract relevant and informative features from network traffic data with a deep learning model incorporating attention mechanisms. The attention mechanism allows the model to focus on the most critical features for accurate intrusion detection. Experimental results on a publicly available IoT intrusion detection dataset demonstrate the effectiveness of the proposed framework in achieving high detection accuracy and low false alarm rates, outperforming existing state-of-the-art methods. The framework offers a robust and adaptable solution for securing IoT networks against evolving cyber threats.
Industrial Control Systems (ICS) are increasingly vulnerable to sophisticated cyberattacks, necessitating robust anomaly detection mechanisms. This paper proposes a novel hybrid deep learning framework for enhanced anomaly detection in high-dimensional ICS data. The framework combines the strengths of Autoencoders (AEs) for feature extraction and dimensionality reduction with Long Short-Term Memory (LSTM) networks for temporal sequence modeling. The AE first learns a compressed representation of normal ICS operational data, effectively capturing the underlying system dynamics. The LSTM network then models the temporal dependencies within the reduced feature space. Anomalies are detected by identifying deviations from the learned normal behavior, leveraging both the reconstruction error of the AE and the prediction error of the LSTM. We evaluate the proposed framework on a benchmark ICS dataset, demonstrating its superior performance compared to state-of-the-art anomaly detection methods in terms of detection accuracy, false positive rate, and robustness to noise. The results highlight the potential of the hybrid approach to significantly improve the security and reliability of critical industrial infrastructure.