文章通过构建扩展长短期记忆网络(Extended Long Short-Term Memory, xLSTM)模型,结合空气质量指数数据和常见天气测量数据,对空气质量指数进行了精确分析以及准确预测。该模型主要由数据预处理、生成数据集和构建模型3个部分组成。在...文章通过构建扩展长短期记忆网络(Extended Long Short-Term Memory, xLSTM)模型,结合空气质量指数数据和常见天气测量数据,对空气质量指数进行了精确分析以及准确预测。该模型主要由数据预处理、生成数据集和构建模型3个部分组成。在数据预处理阶段,文章对常见天气测量数据做数据填充,根据常见天气规律信息和异常值判断算法对常见天气测量数据中的异常值进行数据判断和删除。同时,根据异常值判断算法对空气质量指数数据中的缺失值和异常值进行数据判断和删除。在生成数据集阶段,旧数据集以当前时刻的天气测量数据作为输入、当前时刻的空气质量指数数据作为输出,将旧数据集根据xLSTM模型的输入和输出,转换成当前及之前以一共10个连续时刻的天气测量数据和空气质量指数数据作为输入、下一个时刻的空气质量指数数据作为输出的新数据集。在构建模型阶段,文章构建了xLSTM模型,用来根据当前及之前一共10个连续时刻的天气测量数据和空气质量指数数据预测下一个时刻的空气质量指数数据。实验结果证明,与传统神经网络模型相比,该模型对空气质量指数分析和预测都更加精准。展开更多
The rapid growth of technology impacts all aspects of modern life, including banking and financial transactions. While these industries benefit significantly from technological advancements, they also face challenges ...The rapid growth of technology impacts all aspects of modern life, including banking and financial transactions. While these industries benefit significantly from technological advancements, they also face challenges such as credit card fraud, the most prevalent type of financial fraud. Each year, such fraud leads to billions of dollars in losses for banks, financial institutions, and customers. Although many machine learning (ML) and, more recently, deep learning (DL) solutions have been developed to address this issue, most fail to strike an effective balance between speed and performance. Moreover, the reluctance of financial institutions to disclose their fraud datasets due to reputational risks adds further challenges. This study proposes a predictive model for credit card fraud detection that leverages the unique strengths of Energy-based Restricted Boltzmann Machines (EB-RBM) and Extended Long Short-Term Memory (xLSTM) models. EB-RBM is utilized for its ability to detect new and previously unseen fraudulent patterns, while xLSTM focuses on identifying known fraud types. These models are integrated using an ensemble approach to combine their strengths, achieving a balanced and reliable prediction system. The ensemble employs a bootstrap max-voting mechanism, assigning equal voting rights to EB-RBM and xLSTM, followed by result normalization and aggregation to classify transactions as fraudulent or genuine. The model’s performance is evaluated using metrics such as AUC-ROC, AUC-PR, precision, recall, F1-score, confusion matrix, and elapsed time. Experimental results on a real-world European cardholder dataset demonstrate that the proposed approach effectively balances speed and performance, outperforming recent models in the field.展开更多
文摘文章通过构建扩展长短期记忆网络(Extended Long Short-Term Memory, xLSTM)模型,结合空气质量指数数据和常见天气测量数据,对空气质量指数进行了精确分析以及准确预测。该模型主要由数据预处理、生成数据集和构建模型3个部分组成。在数据预处理阶段,文章对常见天气测量数据做数据填充,根据常见天气规律信息和异常值判断算法对常见天气测量数据中的异常值进行数据判断和删除。同时,根据异常值判断算法对空气质量指数数据中的缺失值和异常值进行数据判断和删除。在生成数据集阶段,旧数据集以当前时刻的天气测量数据作为输入、当前时刻的空气质量指数数据作为输出,将旧数据集根据xLSTM模型的输入和输出,转换成当前及之前以一共10个连续时刻的天气测量数据和空气质量指数数据作为输入、下一个时刻的空气质量指数数据作为输出的新数据集。在构建模型阶段,文章构建了xLSTM模型,用来根据当前及之前一共10个连续时刻的天气测量数据和空气质量指数数据预测下一个时刻的空气质量指数数据。实验结果证明,与传统神经网络模型相比,该模型对空气质量指数分析和预测都更加精准。
文摘The rapid growth of technology impacts all aspects of modern life, including banking and financial transactions. While these industries benefit significantly from technological advancements, they also face challenges such as credit card fraud, the most prevalent type of financial fraud. Each year, such fraud leads to billions of dollars in losses for banks, financial institutions, and customers. Although many machine learning (ML) and, more recently, deep learning (DL) solutions have been developed to address this issue, most fail to strike an effective balance between speed and performance. Moreover, the reluctance of financial institutions to disclose their fraud datasets due to reputational risks adds further challenges. This study proposes a predictive model for credit card fraud detection that leverages the unique strengths of Energy-based Restricted Boltzmann Machines (EB-RBM) and Extended Long Short-Term Memory (xLSTM) models. EB-RBM is utilized for its ability to detect new and previously unseen fraudulent patterns, while xLSTM focuses on identifying known fraud types. These models are integrated using an ensemble approach to combine their strengths, achieving a balanced and reliable prediction system. The ensemble employs a bootstrap max-voting mechanism, assigning equal voting rights to EB-RBM and xLSTM, followed by result normalization and aggregation to classify transactions as fraudulent or genuine. The model’s performance is evaluated using metrics such as AUC-ROC, AUC-PR, precision, recall, F1-score, confusion matrix, and elapsed time. Experimental results on a real-world European cardholder dataset demonstrate that the proposed approach effectively balances speed and performance, outperforming recent models in the field.
