Publication Search

This study discusses a student concentration detection system using Convolutional Neural Network (CNN) with the MobileNetV2 architecture. The dataset was adapted from Classroom Student Behaviors and mapped into four concentration categories: highly focused, focused, less focused, and unfocused. The system was tested with a 720p webcam and produced real-time detection data. The evaluation results show an overall accuracy of 75.85%, with the highest precision achieved in the focused class (0.9859) and the highest recall in the highly focused (0.9739) and unfocused (0.9811) classes. The confusion matrix indicates that the focused class was detected most consistently, while highly focused and unfocused classes were often misclassified as focused, resulting in lower precision. In real-time testing, the system operated at an average of 7 FPS and worked optimally when students faced the camera directly with sufficient lighting, but its performance decreased significantly at face angles greater than 45°. User evaluation shows that 75% of students rated the detection results as accurate/very accurate with an average satisfaction score of 3.6 out of 5, and 75% felt assisted in recognizing their concentration level. From the teachers’ perspective, most stated that the results were consistent with classroom observations, and all expressed willingness to reuse the system.

Social media X (Twitter) has become the main platform for the Indonesian public to express opinions, including on the trend of 'kabur aja dulu' (let's just run away for a bit). This research aims to classify the sentiments of the public using the Naïve Bayes and Support Vector Machine (SVM) methods, and to compare the accuracy of both in sentiment analysis. Data was collected via the Twitter API with the hashtag #kaburajadulu, resulting in 2,067 tweets, which, after the cleansing process and manual labeling, left 385 data points. The analysis process followed the CRISP-DM stages, which include business understanding, data understanding, data preparation, modeling, evaluation, and deployment. Model evaluation was conducted using a confusion matrix with accuracy, precision, and recall metrics. The classification results show that 82% of tweets have a positive sentiment and 18% negative. The Naïve Bayes algorithm achieved an accuracy of 86.49%, slightly lower than SVM, which reached 88.05%. In conclusion, Support Vector Machine is more effective in sentiment classification on public opinion data. This research contributes to the digital mapping of public opinion and recommends the development of automatic labeling methods as well as the exploration of advanced algorithms in the future.

The growth of information technology has driven changes in consumer behavior, one of which is through e-commerce platforms such as Shopee. This phenomenon has generated a large number of customer reviews, including those for local cosmetic products such as Wardah. These reviews serve as an important source of information for understanding customer perceptions and satisfaction levels. However, manual analysis of large and linguistically diverse datasets is inefficient and potentially subjective. This study aims to implement the multi-category Naive Bayes algorithm to classify the sentiment of Wardah product reviews on Shopee into three categories: positive, negative, and neutral. The data were collected using a web scraping technique and processed through a series of preprocessing stages including case folding, tokenization, stopword removal, stemming, and text cleaning. Subsequently, term weighting was performed using the TF-IDF method prior to classification. Model performance was evaluated using a confusion matrix as well as accuracy, precision, and recall metrics. The results indicate that the multi-category Naive Bayes algorithm achieved an accuracy of 86.00%, a precision of 86.63%, and a recall of 98.24%. This approach can assist business practitioners in objectively understanding customer opinions and support decision-making in business strategy and product development.

The rapid development of digital technology has led to an increased adoption of digital payment methods in online transaction-based businesses. However, in practice, failures and limitations in the implementation of digital payment systems still occur, potentially disrupting transaction processes and reducing customer convenience. Payment related obstacles may result in transaction cancellations and increase the risk of customer churn. This study aims to analyze the impact of failures and limitations in digital payment methods on customer churn using a classification-based approach. The data used in this research are secondary e-commerce customer data obtained from the Kaggle platform, including transaction information, payment methods, customer behavior, and historical transaction records. The research methodology consists of data preprocessing, time-based feature engineering, and classification modeling using logistic regression, decision tree, and random forest algorithms. Model performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results indicate that the decision tree model demonstrates superior capability in identifying churn customers compared to the other models, although it does not always achieve the highest accuracy. In addition to digital payment methods, other factors such as purchase value, transaction frequency, purchase timing patterns, and product return rates also influence customer churn. The findings highlight the importance of optimizing digital payment systems as part of customer experience enhancement strategies and customer retention efforts in online transaction–based businesses.

Background: Accurate detection of renal cysts on CT urography requires high diagnostic precision, while manual interpretation by radiologists is susceptible to inter-observer variability and potential delays in clinical decision-making. These challenges underscore the need for a reliable automated detection system to support radiological assessment. Objective: This study aims to develop and evaluate the performance of the Neo-ZasAI application based on the YOLOv8 algorithm for the automatic identification of renal cysts. Methods: Employing a Research and Development design using the ADDIE model, the study encompassed needs analysis, model design, software development, system implementation using 200 CT urography images, and diagnostic performance evaluation. Classification results generated by Neo-ZasAI were compared with radiologist readings through confusion matrix analysis and ROC–AUC assessment. Results: The findings indicate that Neo-ZasAI achieved an accuracy of 97,5%, sensitivity of 96%, specificity of 99%, positive predictive value of 98,9%, and negative predictive value of 96,1%. The ROC analysis yielded an AUC of 0.988 (p < 0.001), demonstrating excellent discriminative capability and high concordance with radiologist interpretations as the diagnostic gold standard. Conclusion: These results suggest that Neo-ZasAI is capable of performing rapid, consistent, and accurate renal cyst detection and is thus feasible for implementation as a clinical decision support system in radiology, with potential integration into PACS workflows and further development to enhance model generalizability.

The increasing complexity and scale of modern network traffic driven by IoT and cloud-based infrastructures have made accurate intrusion detection a critical challenge. Conventional network intrusion detection systems (NIDS) and many deep learning–based approaches struggle to reliably detect minority and stealthy attacks due to severe class imbalance and limited discrimination of subtle traffic patterns. To address these limitations, this study proposes a hybrid CNN–RBF–Attention framework for network intrusion detection. The proposed model integrates three complementary components: (i) a convolutional neural network for hierarchical feature extraction from network flow data, (ii) a radial basis function (RBF) network for localized nonlinear classification using prototype-based decision regions, and (iii) an attention mechanism that adaptively weights RBF activations to emphasize discriminative traffic patterns. SMOTE is applied exclusively to the training data to mitigate class imbalance. The framework is evaluated on the widely used CICIDS2017 and CICIDS2018 benchmark datasets in both binary and multiclass settings, using recall, precision, F1-score, confusion matrices, and ROC analysis. Experimental results demonstrate that the proposed hybrid model consistently outperforms standalone CNN and RBF baselines, particularly in terms of recall and F1-score. On the CICIDS2018 dataset, the model achieves 99.81% accuracy and 99.81% F1-score in binary classification, and 99.54% accuracy and 99.54% F1-score in multiclass classification. On CICIDS2017, it achieves 98.12% accuracy and 98.12% F1-score in binary classification, and 98.92% accuracy and 98.92% F1-score in multiclass classification. Confusion matrix and ROC analyses further show strong class separability and reliable performance in low–false-positive-rate regions, which is critical for real-world IDS deployment. These results confirm that combining deep hierarchical feature learning, localized prototype-based classification, and attention-guided refinement yields a robust, operationally reliable intrusion detection framework for highly imbalanced network environments.

Because it directly impacts revenue, customer loyalty, and long-term business sustainability, customer churn is a critical issue for the e-commerce industry. High churn rates indicate that a business is unable to retain existing customers, which means it is more expensive to acquire new customers. Therefore, a precise analytical approach is needed to identify customer behavior patterns that are likely to churn. Using machine learning methods, this study analyzes and predicts customer churn. For this study, the E-Commerce Customer Churn 2025 dataset, obtained from Kaggle, was used. This dataset consists of 10,000 customer data and contains fifteen variables covering transaction behavior, customer characteristics, and churn status. Data preprocessing, descriptive analysis, exploratory data analysis (EDA), and classification model development using Logistic Regression and Random Forest algorithms were part of the research project. Model evaluation was conducted using a Confusion Matrix and Receiver Operating Characteristic (ROC) Curve to evaluate the model's accuracy and ability to distinguish between churned and non-churned customers. The results showed that the Random Forest model performed better than Logistic Regression, with an ROC-AUC of 1.00. Furthermore, feature importance analysis revealed that the days_since_last_purchase variable was the most dominant factor in predicting customer churn. These findings are expected to help e-commerce companies design more effective, data-driven customer retention strategies.  

The increasing volume of online hotel reviews provides valuable insights into customer perceptions but poses challenges for manual analysis due to its unstructured nature. This study aims to compare the performance of Recurrent Neural Network (RNN) and Bidirectional Encoder Representations from Transformers (BERT) in hotel review sentiment analysis. A total of 20,491 TripAdvisor hotel reviews were classified into three sentiment categories: negative, neutral, and positive. The research methodology includes text preprocessing, stratified data splitting, class imbalance handling using Random Over-Sampling, tokenization, and supervised model training. Model performance was evaluated using a confusion matrix and classification metrics. The results indicate that BERT outperforms RNN, achieving an accuracy of 80.54%, while RNN reached 62.21%. BERT demonstrated superior capability in capturing contextual and semantic information in hotel reviews. These findings suggest that transformer-based models are more effective for sentiment analysis of complex textual data in the hospitality domain and can support data-driven service improvement strategies.    

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.

This study conducts sentiment analysis of Indonesian user reviews of the CapCut application using IndoBERT and compares two evaluation schemes: a single 80/20 train–test split and stratified 5-fold cross-validation (k=5). A total of 1,048,575 reviews were collected from the Google Play Store through web scraping and labeled into three sentiment classes based on rating: negative (1–2), neutral (3), and positive (4–5). After preprocessing—cleaning, case folding, banned-word removal, normalization—and duplicate removal, 517,962 reviews were retained. IndoBERT Base P1 was fine-tuned using fixed hyperparameters (batch size 32, learning rate 2e-5, up to 4 epochs, early stopping patience 2), while undersampling was applied to the training set to address class imbalance. Performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC, supported by confusion matrix and ROC-curve visualizations. The single split achieved an accuracy of 0.756, whereas cross-validation produced a mean accuracy of 0.740. Across both schemes, the positive class achieved the best performance (F1-score 0.850; ROC-AUC 0.918–0.919), while the neutral class remained the most challenging (precision 0.198–0.206; F1-score 0.280–0.283). Overall, cross-validation is recommended for reporting because it reduces dependence on a single partition and provides a more representative estimate across multiple splits.

Chronic Kidney Disease (CKD) is a heterogeneous disorder that gradually affects the structure and function of the kidneys, is difficult to recover, and causes the body to be unable to maintain metabolism and fail to maintain fluid and electrolyte balance, leading to increased urea levels. Chronic kidney disease data was obtained from Kaggle, in this study a comparison was made between two classification algorithms, namely Naïve Bayes Classifier (NBC) and Random Forest because it is not yet known what algorithm is best in classifying chronic kidney disease (CKD). Both algorithms are evaluated based on performance metrics such as accuracy, precision, recall, and confusion matrix. The results of the evaluation showed that in a dataset of 400 samples, the performance  of the Naïve Bayes Classifier (NBC) algorithm obtained an accuracy of 94%, while Random Forest had an accuracy of 93%. Then in the small dataset (158 data), Random Forest got a better accuracy score with 87% compared to the Naïve Bayes Classifier (NBC) of 78%. Based on the results of the evaluation, Random Forest has a more stable performance on small datasets, while Naïve Bayes Classifier (NBC) provides higher performance on larger datasets in the context of chronic kidney disease classification.

The rapid development of digital streaming platforms such as Netflix has generated a large volume of content data with diverse characteristics, thereby requiring effective analytical methods to understand emerging patterns and trends. This study aims to classify Netflix content into two main categories, namely movies and television shows, and to analyze genre trends and content characteristics using a data mining approach with the Naive Bayes algorithm. The dataset used in this study is the Netflix Shows dataset, consisting of 8,809 content entries, with the primary features analyzed including genre, rating, and country of production. The research process begins with data exploration and preprocessing stages, including data cleaning, handling missing values, and transforming categorical features to enable effective model construction. Subsequently, the dataset is divided into training and testing sets to objectively and systematically build and evaluate the Naive Bayes classification model. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics to assess the model’s ability to accurately distinguish between Netflix content types. The experimental results demonstrate that the Naive Bayes algorithm is able to classify Netflix content into Movie and TV Show categories with accuracy, precision, recall, and F1-score values of 100%, respectively. The confusion matrix indicates that no misclassification occurred, suggesting that genre, rating, and country of production features provide a very clear separation between content classes. These findings indicate that the Naive Bayes algorithm can achieve exceptionally high classification performance with optimal evaluation results. The results further reveal distinct differences in characteristics between movies and television shows based on genre and production attributes. Therefore, this study is expected to contribute to the development of content recommendation systems and strategic content management within the streaming industry.

Diabetes Mellitus is a chronic disease that requires early detection to prevent serious complications. This study aims to implement the Artificial Neural Network (ANN) algorithm with the Backpropagation method to predict the risk of diabetes. The dataset used is the Pima Indians Diabetes Dataset, consisting of 768 medical records with 8 feature attributes. This study employs the Multi-Layer Perceptron method with an architecture of 8 input neurons, two hidden layers, and 1 output neuron. Model evaluation is conducted using a Confusion Matrix to measure accuracy levels. The test results show that the model is capable of predicting diabetes diagnosis with an accuracy rate of 76.62%. Based on these results, it can be concluded that the Backpropagation algorithm is effective as an alternative method for early detection of diabetes, although further development is needed to improve the model's sensitivity to positive cases.  

Gestational Diabetes Mellitus (GDM) is a pregnancy-related metabolic disorder that poses health risks to both mother and fetus if not detected early, requiring accurate prediction methods for early screening and clinical decision-making. This study applies the Random Forest algorithm to detect GDM risk using clinical data from the Pima Indian Dataset. Data preprocessing included handling missing values, standardization, feature engineering, and a 70:30 train–test split. Two models were developed: a baseline and an optimized model using GridSearchCV hyperparameter tuning, validated with 5-fold cross-validation. Performance was assessed using a classification report, confusion matrix, and ROC–AUC. Results show that the optimized model outperforms the baseline, achieving 88% accuracy, an AUC of  93%, and average recall of 81%–85%. Compared to previous studies, this approach demonstrates improved predictive performance. The findings indicate that combining Random Forest with comprehensive preprocessing, feature engineering, and model optimization is effective and feasible for developing a medical decision support system for early GDM risk screening.

Indonesian Sign Language (BISINDO) serves as a primary communication medium for the deaf community; however, limited public understanding often creates barriers during daily interactions. This study aims to develop a real-time BISINDO word-level translation system using hand landmark extraction and temporal modeling with Long Short-Term Memory (LSTM). The system employs MediaPipe Hands to detect 21 hand landmarks per frame, which are then processed as sequential motion patterns to classify five BISINDO words: saya, terima kasih, maaf, nama, and kamu. A total of 250 gesture samples were recorded under controlled lighting conditions as the primary dataset. The processed sequences were used to train the LSTM model, which was subsequently integrated with an ESP32 microcontroller and a DFPlayer Mini module to produce direct audio output. Experimental results show that the model achieved an average accuracy of 86%, with precision and recall values ranging from 0.81 to 0.94. The confusion matrix analysis indicates that most gestures were correctly classified, although some errors occurred in gestures with similar initial motion trajectories. Integration testing demonstrated an average system latency of 3.8 seconds and an audio output success rate of 85%. These findings indicate that the proposed system is capable of translating BISINDO word-level gestures accurately, responsively, and consistently in real-time conditions. This study provides a strong foundation for the broader development of sign language translation systems, with potential enhancements in vocabulary expansion, multi-user datasets, and hardware optimization for deployment in real-world environments.

JakLingko is a digital-based public transportation integration system developed to facilitate access to various transportation modes in Jakarta. Along with the increasing number of users, reviews on the JakLingko application reflect user experiences and perceptions. This study aims to analyze the sentiment of user reviews on the Google Play Store using the Naïve Bayes method. Data collection was conducted through web scraping, resulting in 3,260 reviews. The data were preprocessed, sentiment-labeled, and classified using Orange Data Mining. The research applied a quantitative experimental approach with a machine learning framework. The classification results showed that neutral sentiment dominated user reviews, followed by negative and positive sentiments. The Naïve Bayes model achieved 100% accuracy based on the confusion matrix and other evaluation metrics such as precision, recall, and F1-score. The findings highlight that Naïve Bayes can be a reliable approach for analyzing public opinion and serve as a reference for evaluating and improving digital service applications.

Attendance is an essential activity in both educational institutions and companies, serving as an indicator of discipline, presence, and individual responsibility. Conventional attendance systems that still rely on manual journals often face several problems, such as vulnerability to manipulation, data loss, and physical damage. Meanwhile, modern methods such as fingerprint, QR code, RFID, and GPS are not entirely ideal since each has its own limitations in terms of cost, accuracy, user convenience, and potential misuse. For instance, fingerprint systems raise hygiene concerns due to shared use, while QR code and GPS methods are prone to fraud and location spoofing. To address these challenges, this study proposes a face-based attendance simulation system by integrating the YOLOv8 algorithm for face detection and Local Binary Pattern Histogram (LBPH) for face recognition. YOLOv8 was chosen for its ability to detect faces in real time with high speed and accuracy, while LBPH is employed for face recognition due to its robustness in handling variations in facial features and its relatively low computational requirements. This makes the system efficient even when implemented on medium-specification devices. The system was tested on 25 participants with a total of 250 attendance attempts. Based on the confusion matrix analysis, the system achieved outstanding performance with 98.4% accuracy, 98.4% precision, 100% recall, and a 99.2% F1-score. Furthermore, the system automatically recorded attendance dates and times with an average latency of 69.185 ms, proving its capability to operate quickly and reliably in real-world scenarios. Nevertheless, several limitations were observed, such as decreased accuracy when the face moved too quickly during image capture, as well as potential performance degradation under extreme lighting conditions. Despite these challenges, the proposed system demonstrates excellent performance and offers a promising solution for efficient, hygienic, and fraud-resistant attendance management applicable to both educational and professional environments.

The rapid growth of e-commerce platforms has significantly transformed the way consumers share and access product feedback. One of the widely used platforms in Indonesia is Shopee, where customers actively provide reviews of various products, including local skincare brands such as Kahf facial wash. Customer reviews on e-commerce platforms contain valuable information that can be analyzed to understand consumer opinions and preferences. Sentiment analysis, as a branch of natural language processing, enables the classification of textual data into categories such as positive, negative, or neutral. This study aims to classify Shopee user sentiments regarding Kahf facial wash products by implementing the Multinomial Naïve Bayes algorithm, a well-known probabilistic classifier suitable for text categorization. The research methodology consisted of several preprocessing stages, including data cleansing, case folding, tokenizing, stopword removal, and stemming, to prepare raw review texts for further analysis. For feature representation, the Term Frequency–Inverse Document Frequency (TF-IDF) method was applied to capture the importance of words across documents. To evaluate the classification performance, K-Fold cross-validation was employed with K values of 4, 5, 6, and 10 to ensure model reliability and robustness. Considering the issue of imbalanced datasets in user-generated reviews, the Synthetic Minority Over-sampling Technique (SMOTE) was utilized to balance the distribution of sentiment classes. Based on the confusion matrix, the Multinomial Naïve Bayes algorithm demonstrated effective performance in classifying sentiments, achieving satisfactory levels of accuracy, precision, and recall across different folds. These results indicate that the algorithm is capable of handling sentiment analysis tasks for local product reviews effectively. The findings of this study are expected to provide meaningful insights for businesses in understanding consumer perceptions, thereby supporting decision-making processes in product development, marketing strategies, and customer engagement for local brands.

The Paiton Steam Power Plant (PLTU) is one of the main sources of electrical energy in East Java, which plays a vital role in maintaining a sustainable electricity supply. The reliability of generator units is a key element in maintaining stable energy distribution. However, the high frequency of sudden generator failures poses serious challenges, such as increased downtime and increased maintenance costs. To address these challenges, this study aims to design a generator maintenance prediction model based on the Naive Bayes algorithm with a predictive maintenance approach. This study uses historical maintenance data and key sensor parameters such as temperature, oil pressure, and vibration as input. The data is analyzed through several stages, namely data preprocessing, selection of relevant features, and labeling generator conditions into three categories: Normal, Warning, and Critical. The Naive Bayes model is trained to classify the data probabilistically to generate predictions of future generator conditions. Model evaluation using accuracy metrics and a confusion matrix shows that the model successfully achieved an accuracy rate of 89% and was able to provide early warnings of potential failures up to 3 days before failure occurs. The implementation of this system is expected to support the shift in maintenance strategies from reactive and scheduled systems to data-driven predictive systems. Implementing failure predictions allows the technical team at the Paiton PLTU to conduct planned maintenance, avoid sudden disruptions, and extend equipment lifespan. Thus, this model has the potential to reduce operational downtime by up to 25%, while providing significant savings in operational and logistics costs. This research also shows that integrating machine learning technology into energy facility management can improve the efficiency and resilience of the overall electric power system.

This study aims to develop an accurate predictive model for identifying students at risk of academic dropout using Decision Tree and Random Forest algorithms. The research utilizes a publicly available dataset sourced from Kaggle, which includes academic and demographic features such as GPA, attendance, credit load, financial aid status, and exam scores. The methodology involves several stages: data collection, preprocessing (handling missing values, encoding categorical variables, and feature scaling), model training, and evaluation using performance metrics such as Accuracy, Precision, Recall, F1-Score, and Confusion Matrix. Results show that the Random Forest algorithm outperforms Decision Tree in terms of accuracy and robustness, with notable feature importance on math, reading, and writing scores. The findings highlight the potential of machine learning in early detection of dropout risks and provide actionable insights for academic institutions to design timely interventions. This research contributes to the growing field of educational data mining and supports data-driven decision-making processes in higher education management.