Akademik Veri Yönetim Sistemi
International Conference on Artificial Intelligence and Applied Mathematics in Engineering (ICAIAME 2022), Baku, Azerbaycan, 20 - 22 Mayıs 2022, ss.1-3
The long recorded Electrocardiogram signal, and the AF detection method is not effective enough by examining the recording by a specialist clinician. In this study, artificial intelligence-based solutions, which is a method that can solve this problem, are investigated. Although the classical Long Short Term Memory (LSTM) algorithm achieves high accuracy in time series, it has some shortcomings. In order to overcome these shortcomings, a Attention-based block is placed in the LSTM structure, allowing the encoder to control all its hidden states. In addition, the weights of this Attention-based LSTM structure are retrained on an XGBoost classifier, increasing the accuracy. In order to analyze the performance of this designed structure, a public available data set called MIT-BIH AF was used. The dataset has undergone some preprocessing (filtering, feature extraction) before being used in the Attention-based LSTM-XGBoost architecture. It has been proven to be a consistent study with an accuracy rate of 98.9% against studies in the literature. In this way, a supportive decision system has been designed for clinicians and solutions to problems such as computational cost and big data tracking have been presented.