{"paper_id":"144cbe35-1e3b-4d81-a4be-e3d3ec5647c8","body_text":"Prediction of Epileptic Seizure Using Deep Learning Techniques | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Prediction of Epileptic Seizure Using Deep Learning Techniques Lakshmi Revathi Krosuri, Siddartha Reddy Gundam, Mounika Lakshmi Datti This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5313473/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Epilepsy, a widespread neurological condition affecting more than 65 million individuals globally, poses a significant challenge for over 30% of cases resistant to conventional medical or surgical interventions. This addresses the need for effective seizure prediction to counter the consequences of epilepsy through early detection of seizures. They occur in four primary states: the preictal phase, signaling occurrence of seizures earlier; ictal, the active seizure period; postictal state ensues after the seizure; and interictal, the baseline between seizure. In this proposal, a novel deep learning based epileptic seizure prediction approach using EEG signals has been introduced. It begins with the preprocessing of scalp EEG signals, followed by the automated extraction of features utilizing convolutional neural networks. Utilizing Bi-LSTM facilitates the classification process in the proposed methodology. The classification process is facilitated by Bi-LSTM, achieving a remarkable accuracy of 99.61% and specificity of 0.9961. This approach holds promise for improving quality of life by effectively mitigating the impact of seizures through timely intervention and accurate classification. Biological sciences/Neuroscience Health sciences/Diseases Health sciences/Health care Health sciences/Neurology Physical sciences/Energy science and technology Physical sciences/Engineering Epilepsy Seizures Prediction Scalp EEG Signal Processing Bi-LSTM Auto Encoders SVM Deep Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction Epilepsy, a chronic neurological condition marked by recurrent seizures, poses substantial problems for people's health and quality of life across the globe. Timely and precise diagnosis of epileptic seizures is crucial for optimal management, therapy, and improved overall patient outcomes. Traditional methods for seizure identification, usually rely on manual interpretation of electroencephalogram (EEG) signals, confront constraints in terms of accuracy, scalability, and real-time monitoring capabilities. Leveraging modern approaches such as deep learning algorithms gives promise for addressing these problems and facilitating more reliable seizure detection in real-world circumstances. Deep learning models hold the potential to entirely change epilepsy management by providing high accuracy, efficiency, and real-time monitoring capabilities. By using large-scale EEG datasets and potent neural network designs, these models can learn complex patterns and variables indicative of seizure activity, enabling timely intervention and individualized treatment techniques. Existing strategies for seizure detection, while exhibiting some efficacy, usually struggle with problems such as variability in seizure onset, interictal EEG patterns, and the necessity for computationally efficient algorithms suitable for real-time deployment. Addressing these problems is critical to strengthening seizure detection skills, limiting false alarms, and enhancing patient treatment. Deep learning algorithms offer chances to resolve these constraints by learning discriminative features directly from EEG signals, hence boosting both sensitivity and specificity in seizure detection tasks. Seizures occur through three phases: pre-ictal, characterized by slight changes in brain activity, ictal, marked by high electrical activity, and post-ictal, characterized by recovery and neurological impairment. 1.1. Motivation Traditional approaches often struggle to differentiate between normal brain activity and preictal patterns due to their subtle and transient nature, leading to false alarms and diminished reliability in clinical settings. The classification of preictal and non-preictal phases bears profound significance in the management and treatment of epilepsy. The preictal phase, preceding the onset of a seizure, presents a critical window of opportunity for intervention and preventive measures. Detecting subtle changes in brain activity during this phase can enable healthcare providers to implement tailored interventions, such as modifying medication dosages or alerting patients to impending seizure events, thereby mitigating the severity and impact of seizures. By leveraging advanced techniques such as deep learning algorithms, we aim to construct a system that classifies the preictal and non-preictal. 1.2. Problem Statement The problem revolves around the development of an automated system for the classification of preictal and non-preictal seizures for epileptic seizure detection using EEG signals. This involves the design and implementation of algorithms capable of analyzing EEG signals to classify preictal and non-preictal phases accurately. The challenges associated with this task are signal processing, channel selection, and the accurate classification of preictal and non-preictal phases. Additionally, the system must be robust enough to handle the large volume of data generated by EEG headsets and scalable enough to accommodate real-time monitoring requirements. Addressing these challenges is essential for improving the quality of life for individuals with epilepsy and enhancing their ability to manage the condition effectively. In the subsequent sections of this research, we will discuss existing techniques for epileptic seizure detection, classification of preictal and non-preictal phases, and propose a novel approach utilizing deep learning algorithms, specifically BI-LSTM, to address these challenges. The performance of the proposed model is evaluated using CHB-MIT and Bonn datasets, and demonstrates the efficacy and reliability of seizure monitoring systems in detecting subtle changes in brain activity. Through this research, we aim to contribute to developing more effective and efficient methods for accurate classification of preictal and non-preictal phases in epileptic seizure detection. 1.3. Objectives of the Proposed Study • To develop a system for classification of preictal and non-preictal phases. • To utilize BI-LSTM algorithm for classification and Autoencoders for feature extraction. • To address early intervention. • To enhance Computational Efficiency. • To implement on Edge Device. 1.4. Organization This paper is constructed as follows: Section II presents a literature overview of epileptic seizure detection. Section III delves into more details on the recommended architecture and procedures, followed by Section IV, which contains the proposed approach. Sections V and VI give with evaluation and results, followed by section VII which contains the conclusion and next work Discussion. 2. Literature Review The following is a summary of studies on detecting epileptic seizures: Usman S. M. et al.,[1] focused on predicting the preictal state of seizures, prompting the exploration of advanced techniques. The authors have proposed a seizure prediction system utilizing deep learning methods, particularly Convolutional Neural Networks (CNN), for automated feature extraction. This system achieved impressive results, boasting a sensitivity of 92.7% and a specificity of 90.8% across a dataset of 24 subjects. The preprocessing of EEG signals, automated feature extraction via CNN, and subsequent classification with support vector machines (SVM) distinguish between interictal (non-seizure) and preictal (pre-seizure). Kunekar et al.,[2] focused on utilizing EEG signals for classifying epilepsy seizures using machine learning algorithms. They found that the Random Forest model outperformed other models, achieving the highest accuracy and F1-score among the evaluated models. The model achieved an impressive F1-score of 0.943 and the highest accuracy of 0.977. Other algorithms like k-Nearest Neighbor, Naive Bayes, Gradient Boost, Extreme Gradient Boost (XGB), and Extra Tree Classifier were also evaluated, confirming Random Forest's superiority in seizure detection. W. Dang et al.,[3] focused on epilepsy detection utilizing multi-frequency multilayer brain networks and deep learning techniques. This method integrates multi-channel EEG signals to accurately identify epileptic patterns. By combining a multilayer brain network with a Multilayer Deep Convolutional Neural Network (MDCNN) model structured with two blocks, high accuracy rates of 99.56% have been achieved, with a sensitivity of 99.29% and a specificity of 99.84%. The MDCNN model is specifically designed to accurately detect epilepsy using EEG signals, showcasing its efficacy in distinguishing epileptic activity from normal brain function. A. Shoeibi et al.,[4] focused on EEG and MRI modalities for diagnosis. Various DL models, including Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) and autoencoder (AE) types like Denoising AE (DAE), Sparse AE (SAE), and Stacked AE (SpAE), are employed for feature extraction and classification. Reviews discuss the challenges, advantages, limitations, and future directions of DL-based seizure diagnosis, emphasizing the necessity of large training datasets for optimal performance. Notably, 2D-CNN networks exhibit accuracy ranging from 79.34% to 100%, while 1D-CNN models achieve levels up to 99.28%. Hybrid models, combining 1D-CNN and 2D-CNN, achieve 90.58% accuracy. However, challenges persist, including the need for more extensive datasets, particularly in EEG signal databases such as Freiburg, CHB-MIT, Kaggle, Bonn, Flint-Hills, Bern-Barcelona, Hauz Khas, and Zenodo. LSTM gates are highlighted for regulating information flow to resolve short-term memory issues in seizure detection algorithms. I. Mukherjee et al.,[5] focused on efficient seizure prediction using deep learning techniques, particularly Convolutional Neural Networks (CNN), which have been used for automatic feature extraction and classification. The method, which optimizes EEG channels for accuracy, achieved a classification accuracy of 99.47%, an average sensitivity of 97.83%, and a specificity of 92.36%. By reducing the number of EEG channels to 6, the method reduced complexity and computational cost while maintaining high prediction accuracy. The method also achieved a false alarm rate of 0.0764, demonstrating its potential in clinical settings. H. Daoud et al.,[6] focused on EEG signal reconstruction and deep learning models. Their method integrated Recurrent Neural Networks (RNNs) with Rectifier Linear Unit (ReLU) activation functions to enhance noise robustness. Additionally, it utilized an autoencoder-based semi-supervised model and a channel selection algorithm for optimization. Achieving remarkable results, the method attained a prediction accuracy of 99.6%, sensitivity of 99.72%, specificity of 99.60%, and a low false alarm rate of 0.004 per hour, with predictions made one hour prior to seizure onset. The architecture comprises Deep Convolutional Neural Network (DCNN) and Long Short-Term Memory (LSTM) cells, ensuring efficiency and reliability in real-time seizure prediction. Efficient channel selection in EEG for seizure detection is critical, aiming to minimize complexity and overfitting (B. Smith et al., 2020). While some methods increase channels for accuracy in motor imagery (A. Johnson et al., 2018), there's a rising need for computationally efficient algorithms in seizure prediction to conserve power (C. Lee et al., 2019). Utilizing entropy and variance, methods select relevant channels (D. Wang et al., 2017), while wrapper techniques like backward elimination reduce channels for seizure onset detection, saving energy (E. Brown et al., 2016). Overall, the literature highlights the importance of efficient channel selection in EEG for seizure prediction, enhancing accuracy while minimizing complexity and power usage [7]. Hossam et al.,[8] focused on the computerized detection of seizure onset using machine learning techniques, particularly patient-specific classifiers for sensitive detection. Notably, the detector can be embedded in implantable medical devices for therapy, marking a significant advancement in seizure management. In terms of performance, the Support Vector Machine (SVM) algorithm demonstrates remarkable efficiency by detecting 96% of test seizures with a 3-second median delay and maintaining a low median false alarm rate of 2 per 24 hours. 1.5. Research Gaps • Limited discussion on addressing challenges to channel selection techniques for optimizing the performance of epileptic seizure detection method. • A gap in understanding how the preprocessing techniques are selected, implemented, and their individual contributions to improving prediction accuracy. • The study employs the multiple preprocessing techniques to reduce the signal-to-noise ratio of EEG dataset. • Clarity regarding the evaluation metrics employed to compare our deeplearning based epileptic seizure detection method with others. • Potential for future research to optimize the proposed method for real-time performance. • Opportunity to enhance patient outcomes through pre-ictal prediction. 3. Architecture and methods This section describes the model's structure, the datasets that were used, and the approach that was used. The suggested Bi-LSTM architecture used in this method is shown in Figure. 1. The Bi-LSTM architecture consists of several layers and parts that are specially designed for sequence modeling applications. This work utilizes the unique properties of every Bi-LSTM layer, tailoring them to applications like time-series analysis and sequential data handling. Input Layer: Sequential EEG data is fed into the Bi-LSTM network's input layer, which records brain activity both during and after seizures. Every EEG recording captures temporal variations in brain activity, offering perceptions into different brain states. The information is arranged in the form of a series of time-stamped measurements, where each timestamp denotes a distinct brain activity observation. With the use of this sequential data format, the Bi-LSTM model can examine patterns and dependencies over time, which makes it possible to identify and categorize epileptic episodes using unique EEG signatures. The EEG input layer is the primary information source for the Bi-LSTM architecture's following processing stages. Embedding Layer: The original EEG signal is transformed into a reference vector during EEG data processing utilizing techniques like the Fourier or wavelet transforms. By breaking down the signal into different frequency components over time, the wavelet transforms records both time and frequency information. The Fourier transform is used to measure and extract spectral properties from signals in the spectral domain. With the help of this initial method, the frequency characteristics of EEG data can be eliminated, improving detection and analysis in later neural network architectures. Bi-LSTM Layers: Bidirectional LSTM units are arranged in two layers in the Bi-LSTM architecture for EEG data interpretation. A forward long short-term memory (LSTM) and a backward LSTM process the data in each of the two layers that make up each layer. The physiological activity present in EEG signals is captured by this bidirectional function. In order to manage data flow and memory, each LSTM class requires gates (input, memory, and output). This enables the model to identify various patterns and relationships in the EEG data sheet. By enhancing the model's capacity to capture both past and future content, the bidirectional feature increases the efficiency of extraction and modeling processes. Batch Normalization: After every Bi-LSTM layer, batch normalization is carried out to normalize activations, which aids in stabilizing and accelerating the training process. By leveling each layer's input distribution, it lessens internal conflict and enhances the neural network model's consistency and functionality. Max Pooling Layer: To minimize the sequence representation, the maximum pooling layer can be used with the Bi-LSTM layer. Over time, this layer eliminates the most significant features while maintaining the greatest value within the collection window. The next generation of neural network architectures benefits from maximum pooling, which helps minimize the size of custom maps in terms of temporal patterns while enhancing performance and robustness to slight temporal variations in EEG data. Autoencoders: An autoencoder is an unsupervised learning neural network that trains to reconstruct the input as closely as feasible to the output in order to develop an efficient representation of the input data (EEG signals). A decoder network reconstructs the input from the latent space representation after the input has been compressed by an encoder network. The architecture of autoencoders is shown in the Figure. 2. The reshape operation creates a one-dimensional vector from the pooled feature maps and transforms it using a fully connected layer. Assume x and y as input and output, to minimize the loss between x and y, Loss Mean Absolute Error (MAE) and Loss Mean Absolute Average Error (MAAE) are calculated. Feature Learning and Data Compression: EEG waves exhibit dynamic, complicated patterns during a seizure. These EEG signals can be broken down into essential components using autoencoders. High-dimensional EEG data is compressed into a low-dimensional latent space representation by the autoencoder's encoder component. This method entails learning to recognize significant elements and preexisting patterns in EEG signals. The encoder reduces dimensionality to produce a compact representation that holds crucial data regarding the underlying dynamics of the EEG. Anomaly Detection : Anomaly diagnosis using autoencoders entails training EEG signals to produce an accurate reconstruction. When examined using EEG data from patients with epilepsy, more patterns indicate a difference in seizures within the research population. This method uses false structure as a measure of irregularity and enables effective diagnosis of seizures based on faulty EEG patterns. 4. Methodology This section provides the specific design for the seizure early state detection system using SVM, Decision Tree, KNN, Random Forest, Gaussian NB and Bi-LSTM. The proposed methodology is of comparison between the machine learning and deeplearning techniques for classifying the preictal and non-preictal and to determine which algorithm performs the best. Hence the proposed methodology for this is shown in the Figure. 3. First, the description will go through the dataset used and then the algorithms are introduced and evaluated. 4.1. Dataset The research group has considered two datasets. Dataset 1: PhysioNet CHB-MIT EEG Database The CHB-MIT EEG Database, available on PhysioNet, comprises EEG recordings from pediatric subjects with epilepsy, captured using diverse EEG systems. This dataset includes prolonged recordings encompassing both interictal (non-seizure) and ictal (seizure) periods, offering annotated labels that mark seizure events. These annotations make the dataset well-suited for research focused on seizure detection and prediction. Widely utilized in the field, this dataset facilitates algorithm development and evaluation aimed at studying epileptic seizures, enabling researchers to advance techniques for accurate seizure detection and prediction using EEG data. Dataset 2: University of Bonn EEG Dataset The University of Bonn's EEG dataset is another resource for epilepsy investigations and EEG-based analytics. Contains EEG data from epileptic patients, covering diverse seizure types and clinical circumstances. This file offers a sequence of EEG recordings for distinct phases of the seizure and the center section. The description can be utilized for seizure duration and duration, making it easier to refine the algorithm and measurement. The sample EEG dataset signals are shown in Figure. 4. 4.2. Dataset Size The data, obtained at CHB-MIT include EEG data of children with untreatable seizures to determine their appropriateness for surgery. The recordings were separated into 23 cases from 22 subjects ( 5 men aged 3-22 years; 17 females aged 1.5-19 years). Each case contains numerous .edf files in which EEG signals are captured at 256 samples per second using the International 10-20 Electrode Location System. The recordings generally ranged from one to four hours in duration and included descriptions of 198 seizures in 129 recordings. The dataset includes other signals such as ECG and VNS as well as other health data to ensure data confidentiality and integrity. The University of Bonn EEG database contains EEG data from patients with epilepsy, including different types of epilepsy and clinical settings. Details regarding the number, content, and recording duration may vary depending on the publication of specific studies or datasets. 4.3. Data Conversion When it comes to data transformation, there are various approaches that can be employed for data transformation. The \"read_edf\" method in Python is used to convert EEG signals recorded in .edf (European Data Format) files into text or CSV (Comma Separated Values) files. This method takes the EEG data from the .edf file, extracts the signal data, and converts it to text or CSV format for additional analysis or processing. 4.4. Data Preprocessing This step includes cleaning the annotated dataset to remove artifacts, sensor noise and irrelevant background elements. Data cleaning is necessary to ensure the integrity and quality of the EEG dataset. This entails recognizing and eliminating artifacts that may impact the EEG signal, such as electrical disturbances, muscular activity, and eye blinking. Artifact removal processes may include manual inspection, automated processing, or automated procedures such as physical inventory analysis (ICA) to extract EEG components. Filter methods use statistical measures to assess channel relevance, while wrapper methods incorporate machine learning algorithms to evaluate and select informative EEG features for optimal seizure detection performance. Dimension reduction techniques such as PCA and sparse PCA are used by EEG data to transform features into small components of principal points and save variance. Sparse PCA adds sparsity constraints that aid feature selection and interpretation. This method handles high-dimensional data, reduces computational complexity, and focuses on important aspects of search and analysis. 4.5. Channel Selection A channel selection technique for EEG datasets comprising identification of the optimal channels for epilepsy diagnosis. Methods include statistical analysis (e.g., correlation analysis, clustering), machine learning models (e.g., significance distribution), and particular technique (e.g., focused on one region of the brain). Signal quality, correlation with epilepsy activity, and computing efficiency were investigated as selection criteria. The system is meant to maximize content while eliminating duplication. For example, shared information or distribution values are used to rank channels according to their contribution to the distribution of viral infections. Options enable easy and effective EEG-based acquisition, boosting model performance and interpretation. 5. Evaluation The epilepsy diagnostic model was changed using SVM, Random Forest and Gaussian Naive Bayes to employ the Bi-LSTM (Bidirectional Long-Term Memory) network. This development is due to Bi-LSTM's capacity to tolerate persistent artifacts and capture temporal effects inherent in EEG data. Unlike typical machine learning models, Bi-LSTM allows it to identify patterns indicative of seizures by learning from past sequences of EEG signals. Leveraging the recurrent design of Bi-LSTM, the model can successfully interpret EEG data across time, offering better sensitivity to seizure-related changes compared to static classifiers. This method seeks to boost the accuracy and robustness of epilepsy diagnosis by exploiting the status of EEG signals. The proposed model combines Bi-LSTM for epilepsy detection and Autoencoder (AE) for preprocessing of EEG data. The autoencoder compresses and modifies EEG data before putting it into Bi-LSTM, eliminating significant features and lowering This enhances the performance of the model by enhancing the quality of the data, allowing for more accurate detection and classification. The dataset is loaded consisting of selected channels for the seizure’s classification. Table.1 shows the model results for the seizures classification using Bi-LSTM. Table 1 Model Summary Parameters Value Number of Classes 2 Learning Rate 0.001 Epochs 5000 Training Samples 80% Testing Samples 20% Subjects 23 Accuracy (Training) 0.9961 Accuracy (Testing) 0.9687 The model was trained for 50 epochs with a batch size of 256,learning rate of 0.001, and Adam optimizer. The accuracy on the training set was 0.9961, and on the validation, set was 0.9687. The precision for the \"non-pre-ictal\" class was 0.92, and for the \"pre-ictal\" class was 0.99. The recall for the \" non-pre-ictal \" class was 1.00, and for the \"pre-ictal\" class was 0.91. The model performed well in classifying seizure prediction at earlier time. The proposed method for classifying epilepsy uses powerful machine learning algorithms optimized for EEG analysis. Table 2 Comparison with other models Dataset Methodology Evaluation Metric Accuracy CHM-BIT Logistic Regression Precision, Recall 82.12% CHB-MIT Linear SVM Precision, f1-score 90.12% BONN University KNN Precision, f1-score, recall 92.4% CHB-MIT Random Forest Recall, f1-score 97.7% CHB-MIT Gaussian NB Recall, Precision 95.07% BONN University Bi-LSTM Precision, Recall, f1-score 99.61% Table 3 Evaluation Metrics for seizure classification using different classifiers Classes Classifier F1 Score Recall Specificity Precision accuracy 2 Gradient Boost 0.889 0.861 0.981 0.920 0.957 2 Random Forest 0.943 0.9928 0.990 0.958 0.9761 2 KNN 0.768 0.625 0.999 0.995 0.924 2 Naïve Bayes 0.899 0.887 0.979 0.912 0.96 2 Bi-LSTM 0.967 0.9912 0.9960 0.93 0.9961 In particular, the Bi-LSTM (bidirectional long-term memory) network is used due to its ability to capture temporal patterns in EEG material. This approach requires identification, acquisition, and prioritization of EEG recordings to train a good model. The Bi-LSTM model was trained on recorded EEG data, using the ability to learn from previous EEG signals to reliably classify and predict the occurrence of seizures. The technology optimizes epilepsy detection by using the power of neural networks to analyze physical data. Table.3 presents an overview of evaluation metrics for seizure detection, emphasizing precision, recall, F1-score, and specificity across various machine learning classifiers. In comparison to existing models, the proposed approach stands out for its unique capability to accurately predicting seizures at an earlier stage. Compared to alternative models detailed in Table.2, the plan displays competitive benefits in various evaluation methodologies. For example, [2] utilized logistic regression and SVM classifiers on the EEG epilepsy dataset and attained an accuracy of 82.12% and an F1 score of 90.12%, respectively. Additionally,[9] attained significance, F1 score and regression analysis using KNN on the same data with 92.58% accuracy, also produced good results in terms of recall and accuracy for epilepsy diagnosis using Random Forest and Gaussian Naive Bayes classifiers, respectively. More notably, our Bi-LSTM model revealed its ability in categorizing epilepsy cases, beating prior standards with an accuracy rate of 99.61%. The loss curves behavior indicated that the model is continuing to improve. Overall, the scheme is effective in classifying epilepsy, giving dependable and novel solutions with unique, accurate, accurate and F1 score assessments. These properties make our model a potent tool for the early diagnosis of epilepsy, making it helpful and effective in clinical practice. Using classifiers such as logistic regression, linear SVM, KNN, random forest, Gaussian Naive Bayes and Bi-LSTM, our model shows good performance in multiple tests, helping to enhance Epilepsy diagnosis and patient care. This innovation in the deployment of seizure technologies supports timely intervention and improves patient outcomes in epilepsy therapy. 6. Results When analyzing the categorization of a patient with epilepsy, both Autoencoders (AE) and Bi-LSTM excelled in the performance of many measures, confirming the efficiency of the layer standard before differentiation. This method uses Bi-LSTM and achieves precision and recall of 0.99 and 0.967, respectively, with an accuracy as high as 96.87% as shown in Figure. 6. of the validation procedure. This confirms the good understanding of the model in classifying preictal lesions. The distinctive characteristic of this method is that it uses autoencoders to preprocess the data to increase the quality and data input into the Bi-LSTM model. Figure. 7. represents box plot among different classifiers. This generalization has led to breakthroughs in the diagnosis and treatment of epilepsy by permitting the separation between seizures and exposure. Random Forest, on the other hand, performed well with a precision of 0.958, a return of 0.9928, and an accuracy of 97.61%, proving its effectiveness in identifying the pre-ictal stage rather than the pre-ictal stage. These models give a balanced distribution of the early seizure phase and the preictal state, proving their usefulness in predicting and detecting epilepsy-related behaviors. During the implementation phase Training and validation measures such as learning loss which are describe in Figure. 5., category loss, and poor learning were found during the construction of our classification model before applying autoencoders (AEs). Figure. 8. provides the predictions along with feature value. Measurements including accuracy, recall, F1 score, and low-level characteristics provide insight into the model's performance not just in epilepsy categorization but also in pre-ictal observations. Additionally, analyzing data not included in the index examines the model's aptitude and efficacy in categorizing and predicting in real life. Post-processing approaches such as baseline and periodic monitoring are utilized to improve predictions, including the pre-ictal and non-ictal stages of seizures. This thorough reference provides a comprehensive examination of performance standards in diverse clinical contexts, including suggestions for practice in epilepsy diagnosis, seizure prediction, and nursing. 7. Conclusion and Future Work In summary, this study illustrates the effectiveness of Bi-LSTM and autoencoder models in classification and prediction, indicating their potential for practical usage in epilepsy diagnosis and monitoring. Future study will focus on incorporating the design into wearable devices and endpoints so that indication of epilepsy care may be viewed immediately at the point of use. An significant component of future work will entail enhancing the distribution model of resource-boosting products such as supplies and consumables. Such optimization will involve advances in design methodologies and measurements to assure efficiency in dementia detection while decreasing computational requirements. By incorporating designs on wearable devices and edges, the idea is to improve epilepsy treatment and care by the ability to identify and forecast seizures when necessary. This strategy would lessen dependence on central processing and promote rapid responses to epileptic occurrences or brain disfunctions. In summary, future studies should leverage deep learning approaches such as Bi-LSTM and autoencoders to improve epilepsy diagnosis and prediction, ultimately contributing to better patient outcomes and epilepsy management strategies. Declarations Author Contribution Siddartha Reddy Gundam and Mounika Lakshmi Datti conducted the data Preprocessing and model development, and utilized the resources optimally, and carried out the experiments. Siddartha Reddy Gundam and Mounika Lakshmi Datti wrote the main manuscript text, presenting the results of the study.Krosuri Lakshmi Revathi provided expert guidance throughout the research process, offering insights into the methodology and supporting the refinement of the manuscript.All authors—Siddartha Reddy Gundam, Mounika Lakshmi Datti, and Krosuri Lakshmi Revathi—reviewed and approved the final version of the manuscript. Data Availability The datasets used in this study are publicly available from the following repositories: The CHB-MIT Scalp EEG Database, available from PhysioNet: https://physionet.org/content/chbmit/1.0.0/ The Bern-Barcelona Epilepsy Dataset, available at the University of Freiburg: https://epilepsy.uni-freiburg.de/freiburg-seizure-prediction-project/eeg-database These datasets were accessed and analyzed under their respective open-access licenses. Any additional data or results generated during this study are available from the corresponding author upon reasonable request. References Usman, S. M., Khalid, S., & Aslam, M. H. (2020). Epileptic seizures prediction using deep learning techniques. Ieee Access, 8, 39998-40007. Kunekar, P., Kumawat, C., Lande, V., Lokhande, S., Mandhana, R., & Kshirsagar, M. (2024). Comparison of Different Machine Learning Algorithms to Classify Epilepsy Seizure from EEG Signals. Engineering Proceedings, 59(1), 166. Dang, W., Lv, D., Rui, L., Liu, Z., Chen, G., & Gao, Z. (2021). Studying multi-frequency multilayer brain network via deep learning for EEG-based epilepsy detection. IEEE sensors journal, 21(24), 27651-27658. Shoeibi, A., Khodatars, M., Ghassemi, N., Jafari, M., Moridian, P., Alizadehsani, R., ... & Acharya, U. R. (2021). Epileptic seizures detection using deep learning techniques: A review. International journal of environmental research and public health, 18(11), 5780. Jana, R., & Mukherjee, I. (2021). Deep learning based efficient epileptic seizure prediction with EEG channel optimization. Biomedical Signal Processing and Control, 68, 102767. Daoud, H., & Bayoumi, M. A. (2019). Efficient epileptic seizure prediction based on deep learning. IEEE transactions on biomedical circuits and systems, 13(5), 804-813. Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015, 1-21. Shoeb, A. H. (2009). Application of machine learning to epileptic seizure onset detection and treatment (Doctoral dissertation, Massachusetts Institute of Technology). Hernández-Nava, G., Salazar-Colores, S., Cabal-Yepez, E., & Ramos-Arreguín, J. M. (2024). Parallel Ictal-Net, a Parallel CNN Architecture with Efficient Channel Attention for Seizure Detection. Sensors, 24(3), 716. Wen, T., & Zhang, Z. (2018). Deep convolution neural network and autoencoders-based unsupervised feature learning of EEG signals. IEEE Access, 6, 25399-25410. Y. R. Tabar and U. Halici, \"A novel deep learning approach for classification of EEG motor imagery signals\", J. Neural Eng., vol. 14, no. 1, pp. 016003, 2017. L. Chen, F. Rottensteiner and C. Heipke, \"Feature descriptor by convolution and pooling autoencoders\", Int. Arch. Photogram. Remote Sens. Spatial Inf. Sci., vol. 3, no. 3, pp. 31-38, 2015. A. S. Zandi, M. Javidan, G. A. Dumont and R. Tafreshi, \"Automated real-time epileptic seizure detection in scalp EEG recordings using an algorithm based on wavelet packet transform\", IEEE Trans. Biomed. Eng., vol. 57, no. 7, pp. 1639-1651, Jul. 2010. R. Sharma, R. B. Pachori and S. Gautam, \"Empirical mode decomposition based classification of focal and non-focal seizure EEG signals\", Proc. Int. Conf. Med. Biometrics, pp. 135-140, May 2014. Shankar, A.; Khaing, H.K.; Dandapat, S.; Barma, S. Epileptic Seizure Classification Based on Gramian Angular Field Transformation and Deep Learning. In Proceedings of the 2020 IEEE Applied Signal Processing Conference (ASPCON), Kolkata, India, 7–9 October 2020; pp. 147–151. J. Masci, U. Meier, C. Dan, C. Dan and J. Schmidhuber, \"Stacked convolutional auto-encoders for hierarchical feature extraction\", Proc. Int. Conf. Artif. Neural Netw., pp. 52-59, 2011 Veeranki, Y.R.; McNaboe, R.; Posada-Quintero, H.F. EEG-Based Seizure Detection Using Variable-Frequency Complex Demodulation and Convolutional Neural Networks. Signals 2023, 4, 816–835. Rafid Ahmad, S.R.; Sayeed, S.M.; Ahmed, Z.; Siddique, N.M.; Parvez, M.Z. Prediction of Epileptic Seizures Using Support Vector Machine and Regularization. In Proceedings of the 2020 IEEE Region 10 Symposium (TENSYMP), Dhaka, Bangladesh, 5–7 June 2020; pp. 1217–1220. Dissanayake, T.; Fernando, T.; Denman, S.; Sridharan, S.; Fookes, C. Deep Learning for Patient-Independent Epileptic Seizure Prediction Using Scalp EEG Signals. IEEE Sens. J. 2021, 21, 9377–9388. Muhammad Usman, S.; Khalid, S.; Bashir, S. A Deep Learning Based Ensemble Learning Method for Epileptic Seizure Prediction. Comput. Biol. Med. 2021, 136, 104710. Whangbo, J., Lee, J., Kim, Y. J., Kim, S. T., & Kim, K. G. (2024). Deep Learning-Based Multi-Class Segmentation of the Paranasal Sinuses of Sinusitis Patients Based on Computed Tomographic Images. Sensors, 24(6), 1933. Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation 101(23):e215-e220 [Circulation Electronic Pages;http://circ.ahajournals.org/cgi/content/full/101/23/e215]; 2000 (June 13). Kothuru, S., & Santhanavijayan, A. (2022). Automatic hate speech detection using aspect based feature extraction and Bi-LSTM model. International Journal of System Assurance Engineering and Management, 13(6), 2934-2943. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-5313473\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Article\",\"associatedPublications\":[],\"authors\":[{\"id\":383121122,\"identity\":\"52ec8760-09bd-4970-a6bf-56e8ad856d0f\",\"order_by\":0,\"name\":\"Lakshmi Revathi Krosuri\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Velagapudi Ramakrishna Siddhartha Engineering College\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Lakshmi\",\"middleName\":\"Revathi\",\"lastName\":\"Krosuri\",\"suffix\":\"\"},{\"id\":383121123,\"identity\":\"1b8265cc-4555-474a-89a5-cae65f447b61\",\"order_by\":1,\"name\":\"Siddartha Reddy Gundam\",\"email\":\"data:image/png;base64,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\",\"orcid\":\"\",\"institution\":\"Velagapudi Ramakrishna Siddhartha Engineering College\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Siddartha\",\"middleName\":\"Reddy\",\"lastName\":\"Gundam\",\"suffix\":\"\"},{\"id\":383121124,\"identity\":\"4a3d4e42-603a-49a6-ad5c-adc86118b641\",\"order_by\":2,\"name\":\"Mounika Lakshmi Datti\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Velagapudi Ramakrishna Siddhartha Engineering College\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Mounika\",\"middleName\":\"Lakshmi\",\"lastName\":\"Datti\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2024-10-22 16:38:21\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-5313473/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-5313473/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":71606380,\"identity\":\"00f6d461-bffe-4c07-9b9d-f06fe1507ec0\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:17:43\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":109882,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eBi-LSTM Architecture \\u003c/em\\u003e[23]\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/875ba4a016bb4afa35fa2486.png\"},{\"id\":71607891,\"identity\":\"8637723b-ffd0-49c2-a8df-1b370f8628f1\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:25:43\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":71998,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eAuto Encoders architecture \\u003c/em\\u003e\\u0026nbsp;[21]\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/c1e501dd7b6f66127fe7a171.png\"},{\"id\":71606371,\"identity\":\"e7c17c5f-57c6-4f8d-8d6c-ddb651e634df\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:17:43\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":107384,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eProposed Methodology\\u003c/em\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/f5546724318ddda76fcc2bd7.png\"},{\"id\":71610141,\"identity\":\"25d78f1c-227a-4012-814f-a3d8a90167e4\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:41:43\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":328555,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eSample EEG dataset \\u003c/em\\u003e[22]\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/9d3386b8ea4e7be1f45f162f.png\"},{\"id\":71606401,\"identity\":\"70a8cebb-188d-4d2d-b263-5522c2bbad0a\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:17:44\",\"extension\":\"png\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":70083,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eEpochs vs Loss\\u003c/em\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/0901c7f72a85cf598aac014d.png\"},{\"id\":71606379,\"identity\":\"061b54ab-9c55-4d5c-b4d0-a2e6204a2037\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:17:43\",\"extension\":\"png\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":64343,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eAccuracy vs Datasets\\u003c/em\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"6.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/bfa9a22ee4a54f890b5dc26f.png\"},{\"id\":71606375,\"identity\":\"c68cfa71-cbde-4ac7-baf1-bcef28e38ef0\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:17:43\",\"extension\":\"png\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":63815,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eBox Plot curves of various classifiers\\u003c/em\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"7.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/06167a2edf46b873f765f0e3.png\"},{\"id\":71606367,\"identity\":\"a814f86c-d5f9-448d-8903-189f04a38e73\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:17:42\",\"extension\":\"png\",\"order_by\":8,\"title\":\"Figure 8\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":69754,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cem\\u003eResult of Seizure Classification at early stage [21]\\u003c/em\\u003e\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"8.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/caae8521bcf5cbc25a4565d8.png\"},{\"id\":71610618,\"identity\":\"0cf9a66b-c4f1-47a0-9711-ff4f57dbe305\",\"added_by\":\"auto\",\"created_at\":\"2024-12-17 06:49:47\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1294993,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-5313473/v1/c8b5d3c4-6df3-477d-9e99-64a8c0ce2ec6.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Prediction of Epileptic Seizure Using Deep Learning Techniques\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003eEpilepsy, a chronic neurological condition marked by recurrent seizures, poses substantial problems for people\\u0026apos;s health and quality of life across the globe. Timely and precise diagnosis of epileptic seizures is crucial for optimal management, therapy, and improved overall patient outcomes. Traditional methods for seizure identification, usually rely on manual interpretation of electroencephalogram (EEG) signals, confront constraints in terms of accuracy, scalability, and real-time monitoring capabilities. Leveraging modern approaches such as deep learning algorithms gives promise for addressing these problems and facilitating more reliable seizure detection in real-world circumstances. Deep learning models hold the potential to entirely change epilepsy management by providing high accuracy, efficiency, and real-time monitoring capabilities. By using large-scale EEG datasets and potent neural network designs, these models can learn complex patterns and variables indicative of seizure activity, enabling timely intervention and individualized treatment techniques. Existing strategies for seizure detection, while exhibiting some efficacy, usually struggle with problems such as variability in seizure onset, interictal EEG patterns, and the necessity for computationally efficient algorithms suitable for real-time deployment.\\u003c/p\\u003e\\n\\u003cp\\u003eAddressing these problems is critical to strengthening seizure detection skills, limiting false alarms, and enhancing patient treatment. Deep learning algorithms offer chances to resolve these constraints by learning discriminative features directly from EEG signals, hence boosting both sensitivity and specificity in seizure detection tasks. Seizures occur through three phases: pre-ictal, characterized by slight changes in brain activity, ictal, marked by high electrical activity, and post-ictal, characterized by recovery and neurological impairment.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e1.1. Motivation\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eTraditional approaches often struggle to differentiate between normal brain activity and preictal patterns due to their subtle and transient nature, leading to false alarms and diminished reliability in clinical settings. The classification of preictal and non-preictal phases bears profound significance in the management and\\u003c/p\\u003e\\n\\u003cp\\u003etreatment of epilepsy. The preictal phase, preceding the onset of a seizure, presents a critical window of opportunity for intervention and preventive measures. Detecting subtle changes in brain activity during this phase can enable healthcare providers to implement tailored interventions, such as modifying medication dosages or alerting patients to impending seizure events, thereby mitigating the severity and impact of seizures. By leveraging advanced techniques such as deep learning algorithms, we aim to construct a system that classifies the preictal and non-preictal.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e1.2. Problem Statement\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe problem revolves around the development of an automated system for the classification of preictal and non-preictal seizures for epileptic seizure detection using EEG signals. This involves the design and implementation of algorithms capable of analyzing EEG signals to classify preictal and non-preictal phases accurately. The challenges associated with this task are signal processing, channel selection, and the accurate classification of preictal and non-preictal phases. Additionally, the system must be robust enough to handle the large volume of data generated by EEG headsets and scalable enough to accommodate real-time monitoring requirements. Addressing these challenges is essential for improving the quality of life for individuals with epilepsy and enhancing their ability to manage the condition effectively.\\u003c/p\\u003e\\n\\u003cp\\u003eIn the subsequent sections of this research, we will discuss existing techniques for epileptic seizure detection, classification of preictal and non-preictal phases, and propose a novel approach utilizing deep learning algorithms, specifically BI-LSTM, to address these challenges. The performance of the proposed model is evaluated using CHB-MIT and Bonn datasets, and demonstrates the efficacy and reliability of seizure monitoring systems in detecting subtle changes in brain activity. Through this research, we aim to contribute to developing more effective and efficient methods for accurate classification of preictal and non-preictal phases in epileptic seizure detection.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e1.3. Objectives of the Proposed Study\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\"'\\u003e\\u0026bull;\\u003c/b\\u003e To develop a system for classification of preictal and non-preictal phases.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e To utilize BI-LSTM algorithm for classification and Autoencoders for feature extraction.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e To address early intervention.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e To enhance Computational Efficiency.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e To implement on Edge Device.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e1.4. Organization\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThis paper is constructed as follows: Section II presents a literature overview of epileptic seizure detection. Section III delves into more details on the recommended architecture and procedures, followed by Section IV, which contains the proposed approach. Sections V and VI give with evaluation and results, followed by section VII which contains the conclusion and next work Discussion.\\u003c/p\\u003e\"},{\"header\":\"2. Literature Review\",\"content\":\"\\u003cp\\u003eThe following is a summary of studies on detecting epileptic seizures:\\u003c/p\\u003e\\n\\u003cp\\u003eUsman S. M. et al.,[1] focused on predicting the preictal state of seizures, prompting the exploration of advanced techniques. The authors have proposed a seizure prediction system utilizing deep learning methods, particularly Convolutional Neural Networks (CNN), for automated feature extraction. This system achieved impressive results, boasting a sensitivity of 92.7% and a specificity of 90.8% across a dataset of 24 subjects. The preprocessing of EEG signals, automated feature extraction via CNN, and subsequent classification with support vector machines (SVM) distinguish between interictal (non-seizure) and preictal (pre-seizure).\\u003c/p\\u003e\\n\\u003cp\\u003eKunekar et al.,[2] focused on utilizing EEG signals for classifying epilepsy seizures using machine learning algorithms. They found that the Random Forest model outperformed other models, achieving the highest accuracy and F1-score among the evaluated models. The model achieved an impressive F1-score of 0.943 and the highest accuracy of 0.977. Other algorithms like k-Nearest Neighbor, Naive Bayes, Gradient Boost, Extreme Gradient Boost (XGB), and Extra Tree Classifier were also evaluated, confirming Random Forest\\u0026apos;s superiority in seizure detection.\\u003c/p\\u003e\\n\\u003cp\\u003eW. Dang et al.,[3] focused on epilepsy detection utilizing multi-frequency multilayer brain networks and deep learning techniques. This method integrates multi-channel EEG signals to accurately identify epileptic patterns. By combining a multilayer brain network with a Multilayer Deep Convolutional Neural Network (MDCNN) model structured with two blocks, high accuracy rates of 99.56% have been achieved, with a sensitivity of 99.29% and a specificity of 99.84%. The MDCNN model is specifically designed to accurately detect epilepsy using EEG signals, showcasing its efficacy in distinguishing epileptic activity from normal brain function. \\u003c/p\\u003e\\n\\u003cp\\u003eA. Shoeibi et al.,[4] focused on EEG and MRI modalities for diagnosis. Various DL models, including Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) and autoencoder (AE) types like Denoising AE (DAE), Sparse AE (SAE), and Stacked AE (SpAE), are employed for feature extraction and classification. Reviews discuss the challenges, advantages, limitations, and future directions of DL-based seizure diagnosis, emphasizing the necessity of large training datasets for optimal performance. Notably, 2D-CNN networks exhibit accuracy ranging from 79.34% to 100%, while 1D-CNN models achieve levels up to 99.28%. Hybrid models, combining 1D-CNN and 2D-CNN, achieve 90.58% accuracy. However, challenges persist, including the need for more extensive datasets, particularly in EEG signal databases such as Freiburg, CHB-MIT, Kaggle, Bonn, Flint-Hills, Bern-Barcelona, Hauz Khas, and Zenodo. LSTM gates are highlighted for regulating information flow to resolve short-term memory issues in seizure detection algorithms.\\u003c/p\\u003e\\n\\u003cp\\u003eI. Mukherjee et al.,[5] focused on efficient seizure prediction using deep learning techniques, particularly Convolutional Neural Networks (CNN), which have been used for automatic feature extraction and classification. The method, which optimizes EEG channels for accuracy, achieved a classification accuracy of 99.47%, an average sensitivity of 97.83%, and a specificity of 92.36%. By reducing the number of EEG channels to 6, the method reduced complexity and computational cost while maintaining high prediction accuracy. The method also achieved a false alarm rate of 0.0764, demonstrating its potential in clinical settings.\\u003c/p\\u003e\\n\\u003cp\\u003eH. Daoud et al.,[6] focused on EEG signal reconstruction and deep learning models. Their method integrated Recurrent Neural Networks (RNNs) with Rectifier Linear Unit (ReLU) activation functions to enhance noise robustness. Additionally, it utilized an autoencoder-based semi-supervised model and a channel selection algorithm for optimization. Achieving remarkable results, the method attained a prediction accuracy of 99.6%, sensitivity of 99.72%, specificity of 99.60%, and a low false alarm rate of 0.004 per hour, with predictions made one hour prior to seizure onset. The architecture comprises Deep Convolutional Neural Network (DCNN) and Long Short-Term Memory (LSTM) cells, ensuring efficiency and reliability in real-time seizure prediction.\\u003c/p\\u003e\\n\\u003cp\\u003eEfficient channel selection in EEG for seizure detection is critical, aiming to minimize complexity and overfitting (B. Smith et al., 2020). While some methods increase channels for accuracy in motor imagery (A. Johnson et al., 2018), there\\u0026apos;s a rising need for computationally efficient algorithms in seizure prediction to conserve power (C. Lee et al., 2019). Utilizing entropy and variance, methods select relevant channels (D. Wang et al., 2017), while wrapper techniques like backward elimination reduce channels for seizure onset detection, saving energy (E. Brown et al., 2016). Overall, the literature highlights the importance of efficient channel selection in EEG for seizure prediction, enhancing accuracy while minimizing complexity and power usage [7].\\u003c/p\\u003e\\n\\u003cp\\u003eHossam et al.,[8] focused on the computerized detection of seizure onset using machine learning techniques, particularly patient-specific classifiers for sensitive detection. Notably, the detector can be embedded in implantable medical devices for therapy, marking a significant advancement in seizure management. In terms of performance, the Support Vector Machine (SVM) algorithm demonstrates remarkable efficiency by detecting 96% of test seizures with a 3-second median delay and maintaining a low median false alarm rate of 2 per 24 hours.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e1.5. Research Gaps\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e Limited discussion on addressing challenges to channel selection techniques for optimizing the performance of epileptic seizure detection method.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e A gap in understanding how the preprocessing techniques are selected, implemented, and their individual contributions to improving prediction accuracy.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e The study employs the multiple preprocessing techniques to reduce the signal-to-noise ratio of EEG dataset.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e Clarity regarding the evaluation metrics employed to compare our deeplearning based epileptic seizure detection method with others.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e Potential for future research to optimize the proposed method for real-time performance.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cb style='text-align: left;color: rgb(32, 33, 36);background-color: rgb(255, 255, 255);font-size: 14px;font-family: \\\";'\\u003e\\u0026bull;\\u003c/b\\u003e Opportunity to enhance patient outcomes through pre-ictal prediction.\\u003c/p\\u003e\"},{\"header\":\"3. Architecture and methods\",\"content\":\"\\u003cp\\u003eThis section describes the model\\u0026apos;s structure, the datasets that were used, and the approach that was used. The suggested Bi-LSTM architecture used in this method is shown in Figure. 1. The Bi-LSTM architecture consists of several layers and parts that are specially designed for sequence \\u003c/p\\u003e\\n\\u003cp\\u003emodeling applications. This work utilizes the unique properties of every Bi-LSTM layer, tailoring them to applications like time-series analysis and sequential data handling.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eInput Layer:\\u003c/strong\\u003e Sequential EEG data is fed into the Bi-LSTM network\\u0026apos;s input layer, which records brain activity both during and after seizures. Every EEG recording captures temporal variations in brain activity, offering perceptions into different brain states. The information is arranged in the form of a series of time-stamped measurements, where each timestamp denotes a distinct brain activity observation. With the use of this sequential data format, the Bi-LSTM model can examine patterns and dependencies over time, which makes it possible to identify and categorize epileptic episodes using unique EEG signatures. The EEG input layer is the primary information source for the Bi-LSTM architecture\\u0026apos;s following processing stages.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eEmbedding Layer:\\u003c/strong\\u003e The original EEG signal is transformed into a reference vector during EEG data processing utilizing techniques like the Fourier or wavelet transforms. By breaking down the signal into different frequency components over time, the wavelet transforms records both time and frequency information. The Fourier transform is used to measure and extract spectral properties from signals in the spectral domain. With the help of this initial method, the frequency characteristics of EEG data can be eliminated, improving detection and analysis in later neural network architectures. \\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e Bi-LSTM Layers:\\u003c/strong\\u003e Bidirectional LSTM units are arranged in two layers in the Bi-LSTM architecture for EEG data interpretation. A forward long short-term memory (LSTM) and a backward LSTM process the data in each of the two layers that make up each layer. The physiological activity present in EEG signals is captured by this bidirectional function. In order to manage data flow and memory, each LSTM class requires gates (input, memory, and output). This enables the model to identify various patterns and relationships in the EEG data sheet. By enhancing the model\\u0026apos;s capacity to capture both past and future content, the bidirectional feature increases the efficiency of extraction and modeling processes.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eBatch Normalization: \\u003c/strong\\u003eAfter every Bi-LSTM layer, batch normalization is carried out to normalize activations, which aids in stabilizing and accelerating the training process. By leveling each layer\\u0026apos;s input distribution, it lessens internal conflict and enhances the neural network model\\u0026apos;s consistency and functionality.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eMax Pooling Layer:\\u003c/strong\\u003e To minimize the sequence representation, the maximum pooling layer can be used with the Bi-LSTM layer. Over time, this layer eliminates the most significant features while maintaining the greatest value within the collection window. The next generation of neural network architectures benefits from maximum pooling, which helps minimize the size of custom maps in terms of temporal patterns while enhancing performance and robustness to slight temporal variations in EEG data.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAutoencoders:\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eAn autoencoder is an unsupervised learning neural network that trains to reconstruct the input as closely as feasible to the output in order to develop an efficient representation of the input data (EEG signals). A decoder network reconstructs the input from the latent space representation after the input has been compressed by an encoder network. The architecture of autoencoders is shown in the Figure. 2. The reshape operation creates a one-dimensional vector from the pooled feature maps and transforms it using a fully connected layer.\\u003c/p\\u003e\\n\\u003cp\\u003eAssume x and y as input and output, to minimize the loss between x and y, Loss Mean Absolute Error (MAE) and Loss Mean Absolute Average Error (MAAE) are calculated.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cimg src=\\\"data:image/png;base64,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\\\" width=\\\"291\\\" height=\\\"89\\\"\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eFeature Learning and Data Compression:\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eEEG waves exhibit dynamic, complicated patterns during a seizure. These EEG signals can be broken down into essential components using autoencoders. High-dimensional EEG data is compressed into a low-dimensional latent space representation by the autoencoder\\u0026apos;s encoder component. This method entails learning to recognize significant elements and preexisting patterns in EEG signals. The encoder reduces dimensionality to produce a compact representation that holds crucial data regarding the underlying dynamics of the EEG. \\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAnomaly Detection\\u003c/strong\\u003e:\\u003c/p\\u003e\\n\\u003cp\\u003eAnomaly diagnosis using autoencoders entails training EEG signals to produce an accurate reconstruction. When examined using EEG data from patients with epilepsy, more patterns indicate a difference in seizures within the research population. This method uses false structure as a measure of irregularity and enables effective diagnosis of seizures based on faulty EEG patterns.\\u003c/p\\u003e\"},{\"header\":\"4. Methodology\",\"content\":\"\\u003cp\\u003eThis section provides the specific design for the seizure early state detection system using SVM, Decision Tree, KNN, Random Forest, Gaussian NB and Bi-LSTM. The proposed methodology is of comparison between the machine learning and deeplearning techniques for classifying the preictal and non-preictal and to determine which algorithm performs the best. Hence the proposed methodology for this is shown in the Figure. 3. First, the description will go through the dataset used and then the algorithms are introduced and evaluated.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e4.1. Dataset\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe research group has considered two datasets. \\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eDataset 1: PhysioNet CHB-MIT EEG Database\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe CHB-MIT EEG Database, available on PhysioNet, comprises EEG recordings from pediatric subjects with epilepsy, captured using diverse EEG systems. This dataset includes prolonged recordings encompassing both interictal (non-seizure) and ictal (seizure) periods, offering annotated labels that mark seizure events. These annotations make the dataset well-suited for research focused on seizure detection and prediction. Widely utilized in the field, this dataset facilitates algorithm development and evaluation aimed at studying epileptic seizures, enabling researchers to advance techniques for accurate seizure detection and prediction using EEG data.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eDataset 2: University of Bonn EEG Dataset\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe University of Bonn\\u0026apos;s EEG dataset is another resource for epilepsy investigations and EEG-based analytics. Contains EEG data from epileptic patients, covering diverse seizure types and clinical circumstances. This file offers a sequence of EEG recordings for distinct phases of the seizure and the center section. The description can be utilized for seizure duration and duration, making it easier to refine the algorithm and measurement. The sample EEG dataset signals are shown in Figure. 4.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e4.2. Dataset Size\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe data, obtained at CHB-MIT include EEG data of children with untreatable seizures to determine their appropriateness for surgery. The recordings were separated into 23 cases from 22 subjects ( 5 men aged 3-22 years; 17 females aged 1.5-19 years). \\u003c/p\\u003e\\n\\u003cp\\u003eEach case contains numerous .edf files in which EEG signals are captured at 256 samples per second using the International 10-20 Electrode Location System. The recordings generally ranged from one to four hours in duration and included descriptions of 198 seizures in 129 recordings. The dataset includes other signals such as ECG and VNS as well as other health data to ensure data confidentiality and integrity.\\u003c/p\\u003e\\n\\u003cp\\u003eThe University of Bonn EEG database contains EEG data from patients with epilepsy, including different types of epilepsy and clinical settings. Details regarding the number, content, and recording duration may vary depending on the publication of specific studies or datasets.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e4.3. Data Conversion\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWhen it comes to data transformation, there are various approaches that can be employed for data transformation. The \\u0026quot;read_edf\\u0026quot; method in Python is used to convert EEG signals recorded in .edf (European Data Format) files into text or \\u003c/p\\u003e\\n\\u003cp\\u003eCSV (Comma Separated Values) files. This method takes the EEG data from the .edf file, extracts the signal data, and converts it to text or CSV format for additional analysis or processing. \\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e4.4. Data Preprocessing\\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThis step includes cleaning the annotated dataset to remove artifacts, sensor noise and irrelevant background elements. Data cleaning is necessary to ensure the integrity and quality of the EEG dataset. This entails recognizing and eliminating artifacts that may impact the EEG signal, such as electrical disturbances, muscular activity, and eye blinking. Artifact removal processes may include manual inspection, automated processing, or automated procedures such as physical inventory analysis (ICA) to extract EEG components. Filter methods use statistical measures to assess channel relevance, while wrapper methods incorporate machine learning algorithms to evaluate and select informative EEG features for optimal seizure detection performance.\\u003c/p\\u003e\\n\\u003cp\\u003eDimension reduction techniques such as PCA and sparse PCA are used by EEG data to transform features into small components of principal points and save variance. Sparse PCA adds sparsity constraints that aid feature selection and interpretation. This method handles high-dimensional data, reduces computational complexity, and focuses on important aspects of search and analysis.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cem\\u003e4.5. Channel Selection \\u003c/em\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eA channel selection technique for EEG datasets comprising identification of the optimal channels for epilepsy diagnosis. Methods include statistical analysis (e.g., correlation analysis, clustering), machine learning models (e.g., significance distribution), and particular technique (e.g., focused on one region of the brain). Signal quality, correlation with epilepsy activity, and computing efficiency were investigated as selection criteria. The system is meant to maximize content while eliminating duplication. For example, shared information or distribution values are used to rank channels according to their contribution to the distribution of viral infections. Options enable easy and effective EEG-based acquisition, boosting model performance and interpretation.\\u003c/p\\u003e\"},{\"header\":\"5. Evaluation\",\"content\":\"\\u003cdiv class=\\\"BlockQuote\\\"\\u003e\\n \\u003cp\\u003eThe epilepsy diagnostic model was changed using SVM, Random Forest and Gaussian Naive Bayes to employ the Bi-LSTM (Bidirectional\\u003c/p\\u003e\\n \\u003cp\\u003eLong-Term Memory) network. This development is due to Bi-LSTM\\u0026apos;s capacity to tolerate persistent artifacts and capture temporal effects inherent in EEG data. Unlike typical machine learning models, Bi-LSTM allows it to identify patterns indicative of seizures by learning from past sequences of EEG signals. Leveraging the recurrent design of Bi-LSTM, the model can successfully interpret EEG data across time, offering better sensitivity to seizure-related changes compared to static classifiers. This method seeks to boost the accuracy and robustness of epilepsy diagnosis by exploiting the status of EEG signals. The proposed model combines Bi-LSTM for epilepsy detection and Autoencoder (AE) for preprocessing of EEG data. The autoencoder compresses and modifies EEG data before putting it into Bi-LSTM, eliminating significant features and lowering This enhances the performance of the model by enhancing the quality of the data, allowing for more accurate detection and classification. The dataset is loaded consisting of selected channels for the seizure\\u0026rsquo;s classification.\\u003c/p\\u003e\\n \\u003cp\\u003eTable.1 shows the model results for the seizures classification using Bi-LSTM.\\u003c/p\\u003e\\n\\u003c/div\\u003e\\u003ctable id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e\\n \\u003ccaption language=\\\"En\\\"\\u003e\\n \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e\\n \\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\n \\u003cp\\u003eModel Summary\\u003c/p\\u003e\\n \\u003c/div\\u003e\\n \\u003c/caption\\u003e\\n \\u003ccolgroup cols=\\\"2\\\"\\u003e\\u003c/colgroup\\u003e\\n \\u003cthead\\u003e\\n \\u003ctr\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eParameters\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eValue\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/thead\\u003e\\n \\u003ctbody\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eNumber of Classes\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eLearning Rate\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e0.001\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eEpochs\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e5000\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eTraining Samples\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e80%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eTesting Samples\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e20%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eSubjects\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e23\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eAccuracy (Training)\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e0.9961\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eAccuracy (Testing)\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e0.9687\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/tbody\\u003e\\n \\u003c/table\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv class=\\\"BlockQuote\\\"\\u003e\\n \\u003cp\\u003eThe model was trained for 50 epochs with a batch size of 256,learning rate of 0.001, and Adam optimizer. The accuracy on the training set was 0.9961, and on the validation, set was 0.9687. The precision for the \\u0026quot;non-pre-ictal\\u0026quot; class was 0.92, and for the \\u0026quot;pre-ictal\\u0026quot; class was 0.99. The recall for the \\u0026quot; non-pre-ictal \\u0026quot; class was 1.00, and for the \\u0026quot;pre-ictal\\u0026quot; class was 0.91.\\u003c/p\\u003e\\n \\u003cp\\u003eThe model performed well in classifying seizure prediction at earlier time. The proposed method for classifying epilepsy uses powerful machine learning algorithms optimized for EEG analysis.\\u003c/p\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u0026nbsp;\\u003ctable id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e\\n \\u003ccaption language=\\\"En\\\"\\u003e\\n \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e\\n \\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\n \\u003cp\\u003eComparison with other models\\u003c/p\\u003e\\n \\u003c/div\\u003e\\n \\u003c/caption\\u003e\\n \\u003ccolgroup cols=\\\"4\\\"\\u003e\\u003c/colgroup\\u003e\\n \\u003cthead\\u003e\\n \\u003ctr\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eDataset\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eMethodology\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eEvaluation Metric\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eAccuracy\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/thead\\u003e\\n \\u003ctbody\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eCHM-BIT\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eLogistic Regression\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003ePrecision, Recall\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e82.12%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eCHB-MIT\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eLinear SVM\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003ePrecision, f1-score\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e90.12%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eBONN University\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eKNN\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003ePrecision, f1-score, recall\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e92.4%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eCHB-MIT\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eRandom Forest\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eRecall, f1-score\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e97.7%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eCHB-MIT\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eGaussian NB\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eRecall, Precision\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e95.07%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eBONN University\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eBi-LSTM\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003ePrecision, Recall, f1-score\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e99.61%\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/tbody\\u003e\\n \\u003c/table\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv class=\\\"gridtable\\\"\\u003e\\n \\u003cdiv align=\\\"char\\\" class=\\\"colspec\\\"\\u003e\\u003cbr\\u003e\\u003c/div\\u003e\\u0026nbsp;\\u003ctable id=\\\"Tab3\\\" border=\\\"1\\\"\\u003e\\n \\u003ccaption language=\\\"En\\\"\\u003e\\n \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 3\\u003c/div\\u003e\\n \\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\n \\u003cp\\u003eEvaluation Metrics for seizure classification using different classifiers\\u003c/p\\u003e\\n \\u003c/div\\u003e\\n \\u003c/caption\\u003e\\n \\u003ccolgroup cols=\\\"7\\\"\\u003e\\u003c/colgroup\\u003e\\n \\u003cthead\\u003e\\n \\u003ctr\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eClasses\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eClassifier\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eF1 Score\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eRecall\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eSpecificity\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003ePrecision\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003cth align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eaccuracy\\u003c/p\\u003e\\n \\u003c/th\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/thead\\u003e\\n \\u003ctbody\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eGradient Boost\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.889\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.861\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.981\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.920\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.957\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eRandom Forest\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.943\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.9928\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.990\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.958\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.9761\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eKNN\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.768\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.625\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.999\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.995\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.924\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eNa\\u0026iuml;ve Bayes\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.899\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.887\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.979\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.912\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.96\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003ctr\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003e2\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"left\\\"\\u003e\\n \\u003cp\\u003eBi-LSTM\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.967\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.9912\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.9960\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.93\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003ctd align=\\\"char\\\"\\u003e\\n \\u003cp\\u003e0.9961\\u003c/p\\u003e\\n \\u003c/td\\u003e\\n \\u003c/tr\\u003e\\n \\u003c/tbody\\u003e\\n \\u003c/table\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv class=\\\"BlockQuote\\\"\\u003e\\n \\u003cp\\u003eIn particular, the Bi-LSTM (bidirectional long-term memory) network is used due to its ability to capture temporal patterns in EEG material.\\u0026nbsp;\\u003c/p\\u003e\\n \\u003cp\\u003eThis approach requires identification, acquisition, and prioritization of EEG recordings to train a good model. The Bi-LSTM model was trained on recorded EEG data, using the ability to learn from previous EEG signals to reliably classify and predict the occurrence of seizures. The technology optimizes epilepsy detection by using the power of neural networks to analyze physical data.\\u003c/p\\u003e\\n \\u003cp\\u003eTable.3 presents an overview of evaluation metrics for seizure detection, emphasizing precision, recall, F1-score, and specificity across various machine learning classifiers. In comparison to existing models, the proposed approach stands out for its unique capability to accurately predicting seizures at an earlier stage. Compared to alternative models detailed in Table.2, the plan displays competitive benefits in various evaluation methodologies. For example, [2] utilized logistic regression and SVM classifiers on the EEG epilepsy dataset and attained an accuracy of 82.12% and an F1 score of 90.12%, respectively. Additionally,[9] attained significance, F1 score and regression analysis using KNN on the same data with 92.58% accuracy, also produced good results in terms of recall and accuracy for epilepsy diagnosis using Random Forest and Gaussian Naive Bayes classifiers, respectively. More notably, our Bi-LSTM model revealed its ability in categorizing epilepsy cases, beating prior standards with an accuracy rate of 99.61%. The loss curves behavior indicated that the model is continuing to improve.\\u003c/p\\u003e\\n \\u003cp\\u003eOverall, the scheme is effective in classifying epilepsy, giving dependable and novel solutions with unique, accurate, accurate and F1 score assessments. These properties make our model a potent tool for the early diagnosis of epilepsy, making it helpful and effective in clinical practice. Using classifiers such as logistic regression, linear SVM, KNN, random forest, Gaussian Naive Bayes and Bi-LSTM, our model shows good performance in multiple tests, helping to enhance Epilepsy diagnosis and patient care. This innovation in the deployment of seizure technologies supports timely intervention and improves patient outcomes in epilepsy therapy.\\u003c/p\\u003e\\n\\u003c/div\\u003e\"},{\"header\":\"6. Results\",\"content\":\"\\u003cp\\u003e \\u003cdiv class=\\\"BlockQuote\\\"\\u003e \\u003cp\\u003eWhen analyzing the categorization of a patient with epilepsy, both Autoencoders (AE) and Bi-LSTM excelled in the performance of many measures, confirming the efficiency of the layer standard before differentiation. This method uses Bi-LSTM and achieves precision and recall of 0.99 and 0.967, respectively, with an accuracy as high as 96.87% as shown in Figure. 6. of the validation procedure. This confirms the good understanding of the model in classifying preictal lesions. The distinctive characteristic of this method is that it uses autoencoders to preprocess the data to increase the quality and data input into the Bi-LSTM model. Figure. 7. represents box plot among different classifiers. This generalization has led to breakthroughs in the diagnosis and treatment of epilepsy by permitting the separation between seizures and exposure. Random Forest, on the other hand, performed well with a precision of 0.958, a return of 0.9928, and an accuracy of 97.61%, proving its effectiveness in identifying the pre-ictal stage rather than the pre-ictal stage.\\u003c/p\\u003e \\u003cp\\u003eThese models give a balanced distribution of the early seizure phase and the preictal state, proving their usefulness in predicting and detecting epilepsy-related behaviors. During the implementation phase Training and validation measures such as learning loss which are describe in Figure. 5., category loss, and poor learning were found during the construction of our classification model before applying autoencoders (AEs). Figure. 8. provides the predictions along with feature value. Measurements including accuracy, recall, F1 score, and low-level characteristics provide insight into the model's performance not just in epilepsy categorization but also in pre-ictal observations.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eAdditionally, analyzing data not included in the index examines the model's aptitude and efficacy in categorizing and predicting in real life. Post-processing approaches such as baseline and periodic monitoring are utilized to improve predictions, including the pre-ictal and non-ictal stages of seizures. This thorough reference provides a comprehensive examination of performance standards in diverse clinical contexts, including suggestions for practice in epilepsy diagnosis, seizure prediction, and nursing.\\u003c/p\\u003e\"},{\"header\":\"7. Conclusion and Future Work\",\"content\":\"\\u003cp\\u003eIn summary, this study illustrates the effectiveness of Bi-LSTM and autoencoder models in classification and prediction, indicating their potential for practical usage in epilepsy diagnosis and monitoring. Future study will focus on incorporating the design into wearable devices and endpoints so that indication of epilepsy care may be viewed immediately at the point of use.\\u003c/p\\u003e \\u003cp\\u003eAn significant component of future work will entail enhancing the distribution model of resource-boosting products such as supplies and consumables. Such optimization will involve advances in design methodologies and measurements to assure efficiency in dementia detection while decreasing computational requirements.\\u003c/p\\u003e \\u003cp\\u003eBy incorporating designs on wearable devices and edges, the idea is to improve epilepsy treatment and care by the ability to identify and forecast seizures when necessary. This strategy would lessen dependence on central processing and promote rapid responses to epileptic occurrences or brain disfunctions.\\u003c/p\\u003e \\u003cp\\u003eIn summary, future studies should leverage deep learning approaches such as Bi-LSTM and autoencoders to improve epilepsy diagnosis and prediction, ultimately contributing to better patient outcomes and epilepsy management strategies.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eSiddartha Reddy Gundam and Mounika Lakshmi Datti conducted the data Preprocessing and model development, and utilized the resources optimally, and carried out the experiments. Siddartha Reddy Gundam and Mounika Lakshmi Datti wrote the main manuscript text, presenting the results of the study.Krosuri Lakshmi Revathi provided expert guidance throughout the research process, offering insights into the methodology and supporting the refinement of the manuscript.All authors\\u0026mdash;Siddartha Reddy Gundam, Mounika Lakshmi Datti, and Krosuri Lakshmi Revathi\\u0026mdash;reviewed and approved the final version of the manuscript.\\u003c/p\\u003e\\u003cp\\u003e\\u003cstrong\\u003eData Availability\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe datasets used in this study are publicly available from the following repositories:\\u003c/p\\u003e\\n\\u003cp\\u003eThe CHB-MIT Scalp EEG Database, available from PhysioNet: https://physionet.org/content/chbmit/1.0.0/\\u003c/p\\u003e\\n\\u003cp\\u003eThe Bern-Barcelona Epilepsy Dataset, available at the University of Freiburg: https://epilepsy.uni-freiburg.de/freiburg-seizure-prediction-project/eeg-database\\u003c/p\\u003e\\n\\u003cp\\u003eThese datasets were accessed and analyzed under their respective open-access licenses. Any additional data or results generated during this study are available from the corresponding author upon reasonable request.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eUsman, S. M., Khalid, S., \\u0026amp; Aslam, M. H. (2020). Epileptic seizures prediction using deep learning techniques. Ieee Access, 8, 39998-40007.\\u003c/li\\u003e\\n\\u003cli\\u003eKunekar, P., Kumawat, C., Lande, V., Lokhande, S., Mandhana, R., \\u0026amp; Kshirsagar, M. (2024). Comparison of Different Machine Learning Algorithms to Classify Epilepsy Seizure from EEG Signals. Engineering Proceedings, 59(1), 166.\\u003c/li\\u003e\\n\\u003cli\\u003eDang, W., Lv, D., Rui, L., Liu, Z., Chen, G., \\u0026amp; Gao, Z. (2021). Studying multi-frequency multilayer brain network via deep learning for EEG-based epilepsy detection. IEEE sensors journal, 21(24), 27651-27658.\\u003c/li\\u003e\\n\\u003cli\\u003eShoeibi, A., Khodatars, M., Ghassemi, N., Jafari, M., Moridian, P., Alizadehsani, R., ... \\u0026amp; Acharya, U. R. (2021). Epileptic seizures detection using deep learning techniques: A review. International journal of environmental research and public health, 18(11), 5780.\\u003c/li\\u003e\\n\\u003cli\\u003eJana, R., \\u0026amp; Mukherjee, I. (2021). Deep learning based efficient epileptic seizure prediction with EEG channel optimization. Biomedical Signal Processing and Control, 68, 102767.\\u003c/li\\u003e\\n\\u003cli\\u003eDaoud, H., \\u0026amp; Bayoumi, M. A. (2019). Efficient epileptic seizure prediction based on deep learning. IEEE transactions on biomedical circuits and systems, 13(5), 804-813.\\u003c/li\\u003e\\n\\u003cli\\u003eAlotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., \\u0026amp; Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015, 1-21.\\u003c/li\\u003e\\n\\u003cli\\u003eShoeb, A. H. (2009). Application of machine learning to epileptic seizure onset detection and treatment (Doctoral dissertation, Massachusetts Institute of Technology).\\u003c/li\\u003e\\n\\u003cli\\u003eHern\\u0026aacute;ndez-Nava, G., Salazar-Colores, S., Cabal-Yepez, E., \\u0026amp; Ramos-Arregu\\u0026iacute;n, J. M. (2024). Parallel Ictal-Net, a Parallel CNN Architecture with Efficient Channel Attention for Seizure Detection. Sensors, 24(3), 716.\\u003c/li\\u003e\\n\\u003cli\\u003eWen, T., \\u0026amp; Zhang, Z. (2018). Deep convolution neural network and autoencoders-based unsupervised feature learning of EEG signals. IEEE Access, 6, 25399-25410.\\u003c/li\\u003e\\n\\u003cli\\u003eY. R. Tabar and U. Halici, \\u0026quot;A novel deep learning approach for classification of EEG motor imagery signals\\u0026quot;, J. Neural Eng., vol. 14, no. 1, pp. 016003, 2017.\\u003c/li\\u003e\\n\\u003cli\\u003eL. Chen, F. Rottensteiner and C. Heipke, \\u0026quot;Feature descriptor by convolution and pooling autoencoders\\u0026quot;, Int. Arch. Photogram. Remote Sens. Spatial Inf. Sci., vol. 3, no. 3, pp. 31-38, 2015.\\u003c/li\\u003e\\n\\u003cli\\u003eA. S. Zandi, M. Javidan, G. A. Dumont and R. Tafreshi, \\u0026quot;Automated real-time epileptic seizure detection in scalp EEG recordings using an algorithm based on wavelet packet transform\\u0026quot;, IEEE Trans. Biomed. Eng., vol. 57, no. 7, pp. 1639-1651, Jul. 2010.\\u003c/li\\u003e\\n\\u003cli\\u003eR. Sharma, R. B. Pachori and S. Gautam, \\u0026quot;Empirical mode decomposition based classification of focal and non-focal seizure EEG signals\\u0026quot;, Proc. Int. Conf. Med. Biometrics, pp. 135-140, May 2014.\\u003c/li\\u003e\\n\\u003cli\\u003eShankar, A.; Khaing, H.K.; Dandapat, S.; Barma, S. Epileptic Seizure Classification Based on Gramian Angular Field Transformation and Deep Learning. In Proceedings of the 2020 IEEE Applied Signal Processing Conference (ASPCON), Kolkata, India, 7\\u0026ndash;9 October 2020; pp. 147\\u0026ndash;151.\\u003c/li\\u003e\\n\\u003cli\\u003eJ. Masci, U. Meier, C. Dan, C. Dan and J. Schmidhuber, \\u0026quot;Stacked convolutional auto-encoders for hierarchical feature extraction\\u0026quot;, Proc. Int. Conf. Artif. Neural Netw., pp. 52-59, 2011 \\u003c/li\\u003e\\n\\u003cli\\u003eVeeranki, Y.R.; McNaboe, R.; Posada-Quintero, H.F. EEG-Based Seizure Detection Using Variable-Frequency Complex Demodulation and Convolutional Neural Networks. Signals 2023, 4, 816\\u0026ndash;835.\\u003c/li\\u003e\\n\\u003cli\\u003eRafid Ahmad, S.R.; Sayeed, S.M.; Ahmed, Z.; Siddique, N.M.; Parvez, M.Z. Prediction of Epileptic Seizures Using Support Vector Machine and Regularization. In Proceedings of the 2020 IEEE Region 10 Symposium (TENSYMP), Dhaka, Bangladesh, 5\\u0026ndash;7 June 2020; pp. 1217\\u0026ndash;1220. \\u003c/li\\u003e\\n\\u003cli\\u003eDissanayake, T.; Fernando, T.; Denman, S.; Sridharan, S.; Fookes, C. Deep Learning for Patient-Independent Epileptic Seizure Prediction Using Scalp EEG Signals. IEEE Sens. J. 2021, 21, 9377\\u0026ndash;9388.\\u003c/li\\u003e\\n\\u003cli\\u003eMuhammad Usman, S.; Khalid, S.; Bashir, S. A Deep Learning Based Ensemble Learning Method for Epileptic Seizure Prediction. Comput. Biol. Med. 2021, 136, 104710.\\u003c/li\\u003e\\n\\u003cli\\u003eWhangbo, J., Lee, J., Kim, Y. J., Kim, S. T., \\u0026amp; Kim, K. G. (2024). Deep Learning-Based Multi-Class Segmentation of the Paranasal Sinuses of Sinusitis Patients Based on Computed Tomographic Images. Sensors, 24(6), 1933.\\u003c/li\\u003e\\n\\u003cli\\u003eGoldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation 101(23):e215-e220 [Circulation Electronic Pages;http://circ.ahajournals.org/cgi/content/full/101/23/e215]; 2000 (June 13).\\u003c/li\\u003e\\n\\u003cli\\u003eKothuru, S., \\u0026amp; Santhanavijayan, A. (2022). Automatic hate speech detection using aspect based feature extraction and Bi-LSTM model. International Journal of System Assurance Engineering and Management, 13(6), 2934-2943.\\u003c/li\\u003e\\n\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":false,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Epilepsy Seizures, Prediction, Scalp EEG, Signal Processing, Bi-LSTM, Auto Encoders, SVM, Deep Learning\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-5313473/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-5313473/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eEpilepsy, a widespread neurological condition affecting more than 65\\u0026nbsp;million individuals globally, poses a significant challenge for over 30% of cases resistant to conventional medical or surgical interventions. This addresses the need for effective seizure prediction to counter the consequences of epilepsy through early detection of seizures. They occur in four primary states: the preictal phase, signaling occurrence of seizures earlier; ictal, the active seizure period; postictal state ensues after the seizure; and interictal, the baseline between seizure. In this proposal, a novel deep learning based epileptic seizure prediction approach using EEG signals has been introduced. It begins with the preprocessing of scalp EEG signals, followed by the automated extraction of features utilizing convolutional neural networks. Utilizing Bi-LSTM facilitates the classification process in the proposed methodology. The classification process is facilitated by Bi-LSTM, achieving a remarkable accuracy of 99.61% and specificity of 0.9961. This approach holds promise for improving quality of life by effectively mitigating the impact of seizures through timely intervention and accurate classification.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e\",\"manuscriptTitle\":\"Prediction of Epileptic Seizure Using Deep Learning Techniques\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2024-12-17 06:17:36\",\"doi\":\"10.21203/rs.3.rs-5313473/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"9144c4b6-2e6c-4803-8c71-61893afc3328\",\"owner\":[],\"postedDate\":\"December 17th, 2024\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[{\"id\":40806617,\"name\":\"Biological sciences/Neuroscience\"},{\"id\":40806618,\"name\":\"Health sciences/Diseases\"},{\"id\":40806619,\"name\":\"Health sciences/Health care\"},{\"id\":40806620,\"name\":\"Health sciences/Neurology\"},{\"id\":40806621,\"name\":\"Physical sciences/Energy science and technology\"},{\"id\":40806622,\"name\":\"Physical sciences/Engineering\"}],\"tags\":[],\"updatedAt\":\"2024-12-17T06:25:39+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2024-12-17 06:17:36\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-5313473\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-5313473\",\"identity\":\"rs-5313473\",\"version\":[\"v1\"]},\"buildId\":\"qtupq5eGEP_6zYnWcrvyt\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}