Publication Search

Background: The development of industrial technology in the Industry 4.0 era has encouraged the implementation of intelligent monitoring systems to improve machine reliability and operational efficiency. However, machine fault diagnosis systems based on artificial intelligence often face limitations in terms of interpretability because the models used are complex and difficult to explain. Objective: This study aims to develop a deep learning-based industrial machine fault diagnosis system integrated with an Explainable Artificial Intelligence (XAI) approach to improve diagnostic accuracy while providing interpretable insights for users. Method: The research method involves collecting data from industrial machine sensors consisting of vibration signals, temperature measurements, and acoustic signals, followed by data preprocessing and feature extraction processes. The processed data are then used to train a deep learning-based diagnostic model, after which explainability methods such as SHAP or LIME are applied to analyze the contribution of each feature to the model’s prediction results. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics. Results: The results indicate that the proposed deep learning model achieves better performance compared to conventional machine learning methods such as Support Vector Machine and Random Forest. Furthermore, the explainability analysis reveals that vibration amplitude, increases in machine component temperature, and anomalies in acoustic signals are the main factors influencing machine fault detection. Therefore, the proposed system not only improves the accuracy of machine fault diagnosis but also provides transparency in the decision-making process, thereby supporting the implementation of predictive maintenance in smart manufacturing environments.

Budget efficiency is an important issue in state financial management because it is directly related to government spending priorities and their impact on public service programs. Discussions about budget efficiency policies are widespread on social media platform X, generating diverse public responses, thus necessitating an automated approach to understand public opinion trends more quickly and objectively. This research aims to analyze the sentiment of Indonesian people toward budget efficiency policies and compare the performance of the Naïve Bayes and Support Vector Machine (SVM) algorithms in classifying sentiment. The research data used 10,909 Indonesian-language tweets sourced from a public dataset, which were then processed thru the preprocessing stages including cleaning, case folding, normalization, tokenization, stopword removal, and stemming. Sentiment labeling is performed automatically using the Indonesian Sentiment Lexicon (InSet) approach to categorize data into positive, negative, and neutral sentiments. Feature extraction was performed using Term Frequency–Inverse Document Frequency (TF-IDF), and then the data was divided into training and testing sets with an 80:20 ratio. Model performance evaluation was conducted using a confusion matrix and the metrics of accuracy, precision, recall, and F1-score. The research results show that sentiment distribution is dominated by negative sentiment at 56.78%, followed by positive sentiment at 37.40%, and neutral sentiment at 5.83%. In the classification stage, SVM performed best with an accuracy of 86%, while Naïve Bayes achieved an accuracy of 74%. These findings indicate that SVM is more optimal for sentiment classification on social media text data and can be utilized to more effectively support the analysis of public response to budget efficiency policies.

Sentiment analysis is a technique in Natural Language Processing (NLP) used to identify user opinion tendencies based on textual reviews. This study analyzer user reviews of the Maxim application on the Google Play Store and compares three Machine Learning algoritmhs-Naïve Bayes, Support Vector Machine (SVM), and CatBoost-in classifying sentiment. The research stages include data collection, text preprocessing, feature extraction using TF-IDF and Chi-Square, class balancing using SMOTE, and performance evaluation through Accuracy, Precision, Recall, and F1-Score. ANOVA is used to examine the influence of feature selection on model performance. The results show that each model exhibits different performance level across the tested feature combinations. The CatBoost achieved the highest accuracy of 99,26% and demonstrating the most stable performance. Meanwhile, the Naïve Bayes and SVM models experienced performance decreases experiments, especially after applying SMOTE. These findings indicate that the choise of algorithm, feature extraction method, and class balancing technique significantly affects classification outcomes. Overall, CatBoost is identified as the best-performing model, providing more consistenst classification result in accordance with the characteristics of the user reviews.

Sugar consumption in Indonesia remains high, with diabetes affecting 20.4 million people. This condition has prompted the government to introduce an excise policy on Minuman Berpemanis Dalam Kemasan (MBDK) to reduce sugar intake. Social media, particularly the X platform, serves as a medium for the public to express their opinions regarding this policy. This study aims to analyze public sentiment toward the MBDK excise policy using a lexicon-based approach for data labeling and the Multinomial Naive Bayes algorithm with unigram and bigram feature extraction. The initial results show that the highest performance was achieved using 5-Fold Cross Validation, with an average accuracy of 83%, precision of 84%, recall of 75%, and an F1-Score of 77%. After applying data balancing using Stratified Cross Validation combined with Borderline-SMOTE and limiting the features to the 700 most frequent terms, the model’s performance improved. The best results were obtained with 10-Fold Cross Validation, achieving 86% accuracy, 84% precision, 83% recall, and an F1-Score of 83%. These findings indicate that the Multinomial Naive Bayes model can effectively classify public sentiment regarding the MBDK excise policy after the data balancing process.

Application of Multi-Layer Perceptron neural network to fault classification in high-voltage transmission lines is demonstrated in this paper. Different fault types on protected transmission line should be detected and classified rapidly and correctly. This paper presents the use of Discrete Wavelet Transform energy features combined with zero sequence current magnitude as input features for neural network classifier. The proposed method uses eight extracted features to learn hidden relationship in fault signal patterns. Using proposed approach, fault detection and classification of all 11 fault types could be achieved with high accuracy. Improved performance is experienced once the neural network is trained sufficiently with 1188 fault samples, thus performing correctly when faced with different system conditions. Results of performance studies show that proposed neural network-based classifier achieves 96.18% average accuracy, which demonstrates that it can improve the performance of conventional fault classification algorithms, which in turn can provide more efficient solutions in the management and protection of high voltage electrical systems.

Millions of people suffer from malaria, one of the most serious parasitic diseases that threatens human life and causes high rates of morbidity and mortality, particularly in tropical and subtropical regions. Traditional diagnostic methods, such as blood smear examination, which can be performed using a microscope, face many challenges due to the inaccuracy of manual analysis and the reliance on individual skills. Therefore, the use of machine learning or deep learning algorithms to automate malaria detection offers promising solutions to improve accuracy, reduce diagnosis time, and enhance scalability. In this paper, a multi-class convolutional neural network (CNN)-based model is designed to classify cells infected with Plasmodium falciparum (P. falciparum) and Plasmodium vivax (P. vivax) and uninfected cells from blood smears, as most severe cases and deaths are caused by P. falciparum and P. vivax. This is achieved by building and training a CNN from scratch, rather than using transfer learning from pre-trained models. The proposed network was trained and tested on the Kaggle dataset, which consists of 27,558 images of infected and uninfected individuals. These images were divided into 13,779 images of uninfected individuals, 6,890 images of individuals with P. falciparum malaria, and 6,889 images of individuals with P. vivax malaria. The images were preprocessed using several operations, including blurring, denoising, and morphological processing. The proposed model achieved the best evaluation accuracy when compared with other deep learning algorithms, with an accuracy rate of 96.5%, a sensitivity rate of 95%, a specificity rate of 97.6%, and an F1-score rate of 96.5%. These results demonstrate the effectiveness of the proposed model as a tool to assist clinicians in malaria diagnosis, reducing reliance on manual analysis.

Chronic diseases such as chronic kidney disease (CKD), diabetes, and heart disease remain major causes of mortality worldwide, highlighting the need for accurate and interpretable diagnostic models. However, conventional machine learning methods often face challenges of limited generalization, feature redundancy, and class imbalance in medical datasets. This study proposes an integrated classification framework that unifies three complementary feature paradigms: classical tabular attributes, deep latent features extracted through an unsupervised Long Short-Term Memory (LSTM) encoder, and quantum-inspired features derived from a five-qubit circuit implemented in PennyLane. These heterogeneous features are fused using a feature-wise attention mechanism combined with an AdaBoost classifier to dynamically weight feature contributions and enhance decision boundaries. Experiments were conducted on three benchmark medical datasets—CKD, early-stage diabetes, and heart disease—under both balanced and imbalanced configurations using stratified five-fold cross-validation. All preprocessing and feature extraction steps were carefully isolated within each fold to ensure fair evaluation. The proposed hybrid model consistently outperformed conventional and ensemble baselines, achieving peak accuracies of 99.75% (CKD), 96.73% (diabetes), and 91.40% (heart disease) with corresponding ROC AUCs up to 1.00. Ablation analyses confirmed that attention-based fusion substantially improved both accuracy and recall, particularly under imbalanced conditions, while SMOTE contributed minimally once feature-level optimization was applied. Overall, the attention-guided AdaBoost framework provides a robust and interpretable approach for clinical risk prediction, demonstrating that integrating diverse quantum, deep, and classical representations can significantly enhance feature discriminability and model reliability in structured medical data.

This study compares the performance of two deep learning models, namely Convolutional Long Short-Term Memory (ConvLSTM) and Long-term Recurrent Convolutional Network (LRCN), in the task of recognizing human activity from videos. Human activity recognition is an important field in computer vision with many applications, such as security monitoring, human-computer interaction, and social media-based video analysis. ConvLSTM is a model that combines convolution operations with long-term memory LSTM, thus capable of capturing spatial and temporal information simultaneously. This approach is ideal for processing video data sequences that have spatial and temporal dimensions. On the other hand, LRCN combines the power of spatial feature extraction from Convolutional Neural Network (CNN) and temporal sequence modeling through Recurrent Neural Network (RNN), specifically LSTM, to understand movement patterns in videos. The study used the UCF50 dataset consisting of 50 activity classes, but was limited to five classes for the focus of the experiment. The dataset was divided into 80% for training and 20% for testing, and the model was drilled for 50 epochs using early stopping to prevent overfitting. The results show that both models have high training performance. ConvLSTM achieved a training accuracy of around 98% and a validation accuracy of 90%, while LRCN achieved a training accuracy of 99.5% and a validation accuracy of 88%. Although ConvLSTM demonstrated good stability on the validation data, further testing using TikTok videos as real-world data showed that LRCN had a higher confidence level in recognizing activities, with most predictions achieving confidence scores above 80%. This difference in performance indicates that while ConvLSTM excels in generalizing on training data, LRCN is more robust to real-world data variations.

Breast cancer is the most prevalent cancer among women worldwide, requiring early and accurate diagnosis to reduce mortality. This study proposes a hybrid classification pipeline that integrates Hybrid Statistical Feature Selection (HSFS) with unsupervised LSTM-guided feature extraction for breast cancer detection using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. Initially, 20 features were selected using HSFS based on Mutual Information, Chi-square, and Pearson Correlation. To address class imbalance, the training set was balanced using the Synthetic Minority Over-sampling Technique (SMOTE). Subsequently, an LSTM encoder extracted non-linear latent features from the selected features. A fusion strategy was applied by concatenating the statistical and latent features, followed by re-selection of the top 30 features. The final classification was performed using a Support Vector Machine (SVM) with RBF kernel and evaluated using 5-fold cross-validation and a held-out test set. Experimental results showed that the proposed method achieved an average training accuracy of 98.13%, F1-score of 98.13%, and AUC-ROC of 99.55%. On the held-out test set, the model reached an accuracy of 99.30%, precision of 100%, and F1-score of 99.05%, with an AUC-ROC of 0.9973. The proposed pipeline demonstrates improved generalization and interpretability compared to existing methods such as LightGBM-PSO, DHH-GRU, and ensemble deep networks. These results highlight the effectiveness of combining statistical selection and LSTM-based latent feature encoding in a balanced classification framework.

Image classification is a key field in digital image processing with broad applications, such as object recognition and disease detection. The use of artificial neural network architectures, such as MobileNetV2, has significantly advanced pattern recognition in large datasets. However, in small datasets, challenges related to accuracy and generalization are often encountered. This study explores an RGB-based approach utilizing MobileNetV2 for image feature extraction and Support Vector Machine (SVM) as the classifier. MobileNetV2 is applied to extract features from RGB images, which are then further processed by SVM to determine image classes. The results indicate that this model achieves an accuracy of 91.67%, precision of 0.9163, recall of 0.9167, and F1-score of 0.9161. Based on the confusion matrix analysis, the model effectively distinguishes between classes, despite slight overlaps. This research contributes to the development of intelligent image classification systems that can be applied in various fields, including the food industry. With these achievements, the RGB approach integrating MobileNetV2 and SVM has proven effective in enhancing image classification accuracy, even with relatively small datasets. These findings open opportunities for applying similar methods in other image processing tasks that require high accuracy in object or disease detection and classification.

This research focuses on assessing the efficacy of a method that integrates Convolutional Neural Networks (CNN) with Decision Trees for the detection of phishing URLs. Phishing represents a major cyber threat, where cybercriminals attempt to deceive individuals into disclosing sensitive information via fraudulent websites. As the frequency of phishing attacks continues to rise, there is a pressing need for effective detection and prevention strategies. In this investigation, a dataset comprising both phishing and legitimate URLs was utilized to train a CNN-Decision Tree model. The training phase includes feature extraction from URLs using CNN, which excels at identifying intricate patterns within the data, followed by classification through Decision Trees, recognized for their capacity to deliver straightforward and comprehensible interpretations of classification outcomes. The model's performance was evaluated across nine distinct scenarios to assess its effectiveness under varying conditions. The results indicated that the hybrid CNN-Decision Tree model achieved a precision rate of 94%, a recall of 90%, and an F1-Score of 92%, with an overall accuracy of 93%. These findings suggest that the model is not only proficient in identifying phishing URLs but also maintains a commendable balance between precision and recall. This research highlights that the synergy of CNN and Decision Trees can serve as a potent solution for phishing URL detection, significantly contributing to the advancement of enhanced cybersecurity systems.

Due to the variety of types of flowers that exist and having and tracking each variety, making plant lovers and cultivators difficult to distinguish in determining the type of flower, it takes a very long time to find out the type of flower if you only rely on the five senses. With the application of the K-Nearest Neighbor algorithm and feature extraction of color and texture, it is very helpful in image processing to identify flowers more easily and shorten the time, with the greatest accuracy of 71% using the K-7 value, the flower was successfully carried out.

spam is posting unsolicited messages or advertising on social media, particularly Twitter. These messages are normally designed to sell specific products and services or links. In this research, we developed a fuzzy control system to detect Arabic spam tweets based on deep learning with a large language model. Initially, we performed text cleaning and further transformed text into vectors with the help of AraGpt and AraBert. Subsequently, we employed a multi-layer perceptron network model in feature extraction of essential features. Finally, we adopted the fuzzy logic control system for classifying spam tweets using features filtered from deep networks. Employing the proposed Fuzzy logic control system provided nearly a 100% comparative to only utilizing the deep neural networks, which yielded an almost 99% throughput for both large language models Aragpt and Arabert, with a 100% F1 score for the Aragpt model and 99% for Arabert model respectively.

Advances in information and internet technologies have significantly transformed the business environment, including the financial sector. The COVID-19 pandemic has further accelerated this digital adoption, expanding the e-commerce industry and highlighting the necessity for secure online transactions. Credit Card Fraud Detection (CCFD) stands critical as the prevalence of fraudulent activities continues to rise with the increasing volume of online transactions. Traditional methods for detecting fraud, such as rule-based systems and basic machine learning models, tend to fail to keep pace with fraudsters' evolving tactics. This study proposes a novel ensemble deep learning-based approach that combines Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and Multilayer Perceptron (MLP) with the Synthetic Minority Oversampling Technique and Edited Nearest Neighbors (SMOTE-ENN) to address class imbalance and improve detection accuracy. The methodology integrates CNN for feature extraction, GRU for sequential transaction analysis, and Multilayer Perceptron (MLP) as a meta-learner in a stacking framework. By leveraging SMOTE-ENN, the proposed approach enhances data balance and prevents overfitting. With synthetic data, the robustness and accuracy of the model have been improved, particularly in scenarios where fraudulent examples are scarce. The experiments conducted on real-world credit card transaction datasets have established that our approach outperforms existing methods, achieving higher metrics performance.

This study aims to optimize the implementation of the K-Nearest Neighbors (K-NN) algorithm for medical image classification by focusing on selecting the optimal KKK parameter and applying dimensionality reduction techniques to improve accuracy and efficiency. The data used was sourced from public medical image repositories such as The Cancer Imaging Archive (TCIA) and Medical Image Analysis datasets, covering various diseases, including brain tumors, lung cancer, and kidney lesions. The research process involves data collection, data preprocessing, dimensionality reduction using Principal Component Analysis (PCA), applying the K-NN algorithm with Euclidean, Minkowski, and Cosine distance metrics, and performance evaluation using accuracy, precision, recall, and F1-score. Experimental results demonstrate that K=5with the Euclidean distance metric provides the best performance, achieving an accuracy of 90%. Additionally, PCA effectively reduces computational time by 30% without significantly compromising accuracy. This study proves that K-NN is an effective method for medical image classification. However, further research is needed to integrate K-NN with deep learning models to enhance performance and feature extraction capabilities.

Skin cancer (SC) is a highly serious kind of cancer that, if not addressed swiftly, might result in the patient’s demise. Early detection of this condition allows for more effective therapy and prevents disease development. Deep Learning (DL) approaches may be used as an effective and efficient tool for SC detection (SCD). Several DL-based algorithms for automated SCD have been reported. However, more efficient models are needed to improve accuracy. As a result, this paper introduces a new strategy for SCD based on Grey Wolf optimization (GWO) methodologies and CNN. The proposed methodology has four stages: preprocessing, segmentation, feature extraction, and classification. The proposed method utilizes a Convolutional Neural Network (CNN) to extract features from Regions of Interest (ROIs). CNN is employed for feature categorization, whereas the GWO approach enhances accuracy by refining edge detection and segmentation. This technique utilizes a probabilistic model to accelerate the convergence of the GWO algorithm. Employing the GWO model to optimize the structure and weight vectors of CNNs can enhance diagnostic accuracy by a minimum of 5%, based on evaluation outcomes. The application of the proposed strategy and its performance comparison with other methods indicate that the proposed method with GWO predicted SC with an average accuracy of 95.11% and without GWO an Accuracy of 92.66%, respectively, enhancing accuracy by a minimum of 2.5% when we train our model with GWO.

Signature verification is crucial for confirming the authenticity of identities in both administrative and financial transactions, where signature forgery can lead to significant security risks. The Harris Corner Detector Algorithm is a widely used method for feature extraction in image processing; its application spans various domains, such as detection of signature forgery. While effective in identifying key features, noise significantly affects performance, especially with impulse noise like salt-and-pepper noise commonly found in signature images. To solve this problem, this study enhances the Harris Corner Detector Algorithm by applying a median filter before gradient calculation. This method removes noise without sacrificing the integrity of key features important in signature forgery detection. The study evaluates the original and the enhanced algorithm using standard image quality metrics. Peak Signal-to-Noise Ratio (PSNR) surged from an average of 13.6 dB to 43.28 dB, the Structural Similarity Index (SSIM) improved significantly from 78% to 94%, and the Mean Squared Error (MSE) dropped substantially from 16.74 to 3.84. These advancements resulted in a more reliable algorithm, exhibiting excellent resistance to noise while maintaining image structure, making the enhanced algorithm highly effective for accurate signature forgery detection.