Garuda - Garba Rujukan Digital

Stacked LSTM with Multi Head Attention Based Model for Intrusion Detection S Phani Praveen; Padmavathi Panguluri; Uddagiri Sirisha; Deshinta Arrova Dewi; Tri Basuki Kurniawan; Lusiana Efrizoni
Journal of Applied Data Sciences Vol 7, No 1: January 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i1.764

The rapid advancement of digital technologies, including the Internet of Things (IoT), cloud computing, and mobile communications, has intensified reliance on interconnected networks, thereby increasing exposure to diverse cyber threats. Intrusion Detection Systems (IDS) are essential for identifying and mitigating these threats; however, traditional signature-based and rule-based methods fail to detect unknown or complex attacks and often generate high false positive rates. Recent studies have explored machine learning (ML) and deep learning (DL) approaches for IDS development, yet many suffer from poor generalization, limited scalability, and an inability to capture both spatial and temporal dependencies in network traffic. To overcome these challenges, this study proposes a hybrid deep learning framework integrating Convolutional Neural Networks (CNN), Stacked Long Short-Term Memory (LSTM) networks, and a Multi-Head Self-Attention (MHSA) mechanism. CNN layers extract spatial features, stacked LSTM layers capture long-term temporal dependencies, and MHSA enhances focus on the most relevant time steps, improving accuracy and reducing false alarms. The proposed model was trained and evaluated on the UNSW-NB15 dataset, which represents modern attack vectors and realistic network behavior. Experimental results show that the model achieves state-of-the-art performance, attaining 99.99% accuracy and outperforming existing ML and DL-based intrusion detection systems in both precision and generalization capability.

Incorporate Transformer-Based Models for Anomaly Detection Deshinta Arrova Dewi; Harprith Kaur Rajinder Singh; Jeyarani Periasamy; Tri Basuki Kurniawan; Henderi Henderi; M. Said Hasibuan; Yogeswaran Nathan
Journal of Applied Data Sciences Vol 6, No 3: September 2025
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v6i3.762

This paper explores the effectiveness of Transformer-based models, specifically the Time-Series Transformer (TST) and Temporal Fusion Transformer (TFT), for anomaly detection in streaming data. We review related work on anomaly detection models, highlighting traditional methods' limitations in speed, accuracy, and scalability. While LSTM Autoencoders are known for their ability to capture temporal patterns, they suffer from high memory consumption and slower inference times. Though efficient in terms of memory usage, the Matrix Profile provides lower performance in detecting anomalies. To address these challenges, we propose using Transformer-based models, which leverage the self-attention mechanism to capture long-range dependencies in data, process sequences in parallel, and achieve superior performance in both accuracy and efficiency. Our experiments show that TFT outperforms the other models with an F1-score of 0.92 and a Precision-Recall AUC of 0.71, demonstrating significant improvements in anomaly detection. The TST model also shows competitive performance with an F1-score of 0.88 and Precision-Recall AUC of 0.68, offering a more efficient alternative to LSTMs. The results underscore that Transformer models, particularly TST and TFT, provide a robust solution for anomaly detection in real-time applications, offering improved performance, faster inference times, and lower memory usage than traditional models. In conclusion, Transformer-based models stand out as the most effective and scalable solution for large-scale, real-time anomaly detection in streaming time-series data, paving the way for their broader application across various industries. Future work will further focus on optimizing these models and exploring hybrid approaches to enhance detection capabilities and real-time performance.

Detecting Gender-Based Violence Discourse Using Deep Learning: A CNN-LSTM Hybrid Model Approach Tri Basuki Kurniawan; Deshinta Arrova Dewi; Henderi Henderi; M. Said Hasibuan; Mohd Zaki Zakaria; Abdul Azim Bin Ismail
Journal of Applied Data Sciences Vol 6, No 3: September 2025
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v6i3.761

Gender-Based Violence (GBV) is a critical social issue impacting millions worldwide. Social media discussions offer valuable insights into public awareness, sentiment, and advocacy, yet manually analyzing such vast textual data is highly challenging. Traditional text classification methods often struggle with contextual understanding and multi-class categorization, making it difficult to accurately identify discussions on Sexual Violence, Physical Violence, and other topics. To address this, the present study proposes a hybrid deep learning approach combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. CNN is utilized for extracting key linguistic features, while LSTM enhances the classification process by maintaining sequential dependencies. This hybrid CNN+LSTM model is evaluated against standalone CNN and LSTM models to assess its performance in classifying GBV-related tweets. The dataset was sourced from Kaggle, containing real-world Twitter discussions on GBV. Experimental results demonstrate that the hybrid model surpasses both CNN and LSTM models, achieving an accuracy of 89.6%, precision of 88.4%, recall of 89.1%, and F1-score of 88.7%. Confusion matrix and ROC curve analyses further confirm the hybrid model’s superior performance, correctly identifying Sexual Violence (82%), Physical Violence (15%), and Other (3%) cases with reduced misclassification rates. These results suggest that combining CNN’s feature extraction with LSTM’s contextual learning provides a more balanced and effective classification model for GBV-related text. This work supports the development of AI-based tools for social media monitoring, policy-making, and advocacy, helping stakeholders better understand and respond to GBV discussions. Future research could explore transformer-based models like BERT and real-time classification applications to further improve performance.

Navigating Heart Stroke Terrain: A Cutting-Edge Feed-Forward Neural Network Expedition S Phani Praveen; Jeevana Sujitha Mantena; Uddagiri Sirisha; Deshinta Arrova Dewi; Tri Basuki Kurniawan; Choo Wou Onn; Yorman Yorman
Journal of Applied Data Sciences Vol 6, No 3: September 2025
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v6i3.763

Heart stroke remains one of the leading causes of death worldwide, necessitating early and accurate prediction systems to enable timely medical intervention. While a variety of machine learning approaches have been employed to address this issue, including Logistic Regression, Decision Trees, Random Forests, Support Vector Machines, and K-Nearest Neighbors, these models often suffer from limitations such as overfitting, insufficient generalization, poor performance on imbalanced datasets, and inability to capture complex nonlinear patterns in clinical data. Additionally, many existing works do not comprehensively integrate both clinical and demographic features or lack rigorous evaluation metrics beyond accuracy alone. This study proposes a novel Feed-Forward Neural Network (FFNN) model for heart stroke prediction, designed to overcome the shortcomings of conventional models. Unlike shallow classifiers, the FFNN architecture employed here leverages multiple hidden layers and nonlinear activation functions to learn intricate relationships within the dataset. The dataset used comprises various attributes such as age, hypertension, heart disease, BMI, and smoking status, which were preprocessed through normalization, one-hot encoding, and imputation techniques to ensure data quality and model performance. Experiments were conducted using a stratified train-test split, and the model was trained using the Adam optimizer with carefully tuned hyperparameters. Comparative evaluations against baseline models (Logistic Regression, Random Forest, and SVM) were carried out using precision, recall, F1-score, and ROC-AUC as performance metrics. The proposed FFNN achieved the highest accuracy of 96.47%, along with substantial improvements in recall and F1-score, highlighting its superior capability in identifying potential stroke cases even in imbalanced datasets. This work bridges a significant gap in heart stroke prediction by demonstrating the effectiveness of deep learning models—specifically FFNNs—in extracting complex patterns from diverse patient data. It also sets the stage for further exploration of deep learning-based clinical decision support systems.

Stacked LSTM with Multi Head Attention Based Model for Intrusion Detection S Phani Praveen; Padmavathi Panguluri; Uddagiri Sirisha; Deshinta Arrova Dewi; Tri Basuki Kurniawan; Lusiana Efrizoni
Journal of Applied Data Sciences Vol 7, No 1: January 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i1.764

The rapid advancement of digital technologies, including the Internet of Things (IoT), cloud computing, and mobile communications, has intensified reliance on interconnected networks, thereby increasing exposure to diverse cyber threats. Intrusion Detection Systems (IDS) are essential for identifying and mitigating these threats; however, traditional signature-based and rule-based methods fail to detect unknown or complex attacks and often generate high false positive rates. Recent studies have explored machine learning (ML) and deep learning (DL) approaches for IDS development, yet many suffer from poor generalization, limited scalability, and an inability to capture both spatial and temporal dependencies in network traffic. To overcome these challenges, this study proposes a hybrid deep learning framework integrating Convolutional Neural Networks (CNN), Stacked Long Short-Term Memory (LSTM) networks, and a Multi-Head Self-Attention (MHSA) mechanism. CNN layers extract spatial features, stacked LSTM layers capture long-term temporal dependencies, and MHSA enhances focus on the most relevant time steps, improving accuracy and reducing false alarms. The proposed model was trained and evaluated on the UNSW-NB15 dataset, which represents modern attack vectors and realistic network behavior. Experimental results show that the model achieves state-of-the-art performance, attaining 99.99% accuracy and outperforming existing ML and DL-based intrusion detection systems in both precision and generalization capability.

Title

Found 5 Documents
Search

Abstract

Abstract

Abstract

Abstract

Abstract

Title Search

Found 5 Documents Search

Abstract

Abstract

Abstract

Abstract

Abstract

Title

Found 5 Documents
Search