Publication Search

Nature tourism is one of the sectors that plays an important role in supporting the development of regional tourism, including in Lahat Regency, which has significant waterfall tourism potential. Currently, many visitors share their reviews and experiences through digital platforms such as Google Maps. This review can be used as a source of information to understand the public's evaluation of the quality of tourist attractions. This study aims to examine public perception of tourist attractions in Lahat Regency using the Support Vector Machine (SVM) method. Research data were collected through scraping from Google Maps, totaling 500 reviews from five tourist attractions, namely Curup Maung, Curup Buluh, Senyawe Waterfall, Panjang Waterfall, and Green Canyon. The research stages include data preprocessing, consisting of cleaning, case folding, normalization, tokenization, stopword removal, and stemming. After that, feature extraction was carried out using the TF-IDF method and the classification process using the SVM algorithm. Based on the research results, the Support Vector Machine (SVM) method is able to perform sentiment classification quite well, although the accuracy level varies for each tourist attraction. Curup Maung and Panjang Waterfall achieved the highest accuracy level of 90%. Nevertheless, most visitor reviews were dominated by negative sentiments. This indicates that there are still several aspects that need to be improved, particularly related to tourist facilities and services. This research is expected to serve as a consideration for tourism managers and local governments in efforts to improve management quality as well as the development of tourism in Lahat Regency.

Sign language recognition is one of the areas of image recognition and image processing technology that is developing rapidly in human-computer interaction. This technology really helps the deaf and speech impaired in communicating with non-disabled people. This research aims to examine the optimization of an object tracking system in sign language using the Gaussian Mixture Model (GMM) and Kalman Filter by including the Region of Interest (ROI). The proposed system consists of three main components, namely hand detection, object extraction, and classification. Hand detection is done using the Kalman Filter to track hand movements accurately. Next, Region of Interest (ROI) features, such as shape, direction and movement features, are extracted from the detected part of the hand. These features are fed into a Gaussian Mixture Model (GMM) classifier, which can recognize sign language based on the extracted features. With the combination of GMM and Kalman Filter in this research, it can increase accuracy in object tracking, reduce interference from the background, and ensure the tracking focus remains on important objects. The dataset used is in the form os SIBI alphabet symbols, namely A-Z with the amount of data for each class, namely 620 images. Based on the research result, model testing using GMM, Kalman Filter and ROI produces higher accuracy of 99%, while model testing using GMM and ROI produces accuracy of 90%.

The rise of social media has created a digital public sphere that enables users to express their opinions on social and political issues openly and in real-time. One of the most discussed topics on social media platform X is the trending hashtag #IndonesiaGelap, which reflects public concern and criticism regarding various governmental and societal conditions. This study aims to conduct sentiment analysis on tweets containing the hashtag to determine the overall sentiment trend among users. The method employed in this research is the Naive Bayes classification algorithm, known for its simplicity and effectiveness in text classification. To enhance the model’s performance, Particle Swarm Optimization (PSO) is applied to optimize feature selection and parameter tuning. The dataset consists of public tweets collected via the Twitter API, followed by preprocessing, feature extraction using TF-IDF, and sentiment classification into three categories: positive, negative, and neutral. The results indicate that the integration of PSO significantly improves the classification accuracy of the Naive Bayes model compared to the baseline. The majority of tweets related to #IndonesiaGelap exhibit a negative sentiment, indicating widespread public dissatisfaction and criticism. This research is expected to contribute to a better understanding of public perception and serve as valuable input for stakeholders in addressing social issues in the digital age.

Communication is an essential human right, yet a significant communication gap persists between individuals with sensory disabilities, specifically the deaf and speech-impaired, and the general public. While many technological solutions have been proposed to translate sign language, existing models primarily rely on heavy deep learning architectures such as Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN/LSTM). These models often demand high computational power, leading to latency and limiting real-time application on standard devices. This study proposes a lightweight, fast, and highly responsive sign language translation system specifically designed to recognize static alphabets (A-Z) and single-character air writing. The system utilizes MediaPipe for hand tracking, where feature extraction is intelligently processed by calculating the relative spatial coordinates of fingertips to the wrist, reducing dependency on raw camera coordinates. Classification is performed using a Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel, prioritizing computational efficiency without sacrificing accuracy. To enhance user experience, the system introduces three key novelties: smart relative feature extraction, an anti-duplication hold system with a 1-second timer to prevent input spamming, and a non-blocking multithreaded audio execution (Daemon Thread) utilizing Google Text-to-Speech (gTTS), ensuring the webcam feed remains fluid during audio playback. Additionally, an alternative air-writing mode is integrated, utilizing geometric heuristics and PyTesseract OCR to read single drawn letters in the air. The results indicate that the proposed system operates swiftly and efficiently, bridging the communication barrier with a hardware-friendly approach.

The integration of Artificial Intelligence (AI) into precision agriculture has significantly improved plant disease recognition; however, many existing deep learning models remain computationally expensive and feature-redundant, limiting their deployment on low-power and edge devices. To address these limitations, this study proposes a lightweight framework for maize leaf disease recognition based on serial deep feature extraction, dimensionality reduction, and machine-learning–based classification. A pre-trained MobileNetV2 network is employed as a fixed feature extractor to obtain discriminative visual representations, while Principal Component Analysis (PCA) is applied to reduce feature dimensionality by approximately 76%, retaining 95% of the original variance and improving computational efficiency. The compressed features are subsequently classified using a Radial Basis Function Support Vector Machine (RBF-SVM), optimized via grid search and cross-validation. Experiments conducted on a four-class maize leaf disease dataset (Northern Leaf Blight, Common Rust, Gray Leaf Spot, and Healthy), with class imbalance handled during training, demonstrate that the proposed MobileNetV2–PCA–SVM pipeline achieves 97.58% accuracy, 96.60% precision, 96.59% recall, and 96.59% F1-score, outperforming the DenseNet201 + Bayesian-optimized SVM baseline (94.60%, 94.40%, 94.40%, and 94.40%, respectively). This improvement corresponds to a 2.98% accuracy gain, a 55% reduction in error rate, an 86% reduction in model parameters (20.31M to 2.75M), and an 85% reduction in model size (81 MB to 12 MB). These results indicate that the proposed framework provides a compact and efficient solution with strong potential for deployment in resource-constrained agricultural environments.

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.

Since the announcement of the policy plan for a name transfer system in the sale of used mobile phones, the issue has attracted widespread public attention and discussion. People have expressed their opinions on social media platforms, particularly TikTok. This study aims to classify the sentiment of TikTok users using Naive Bayes and Support Vector Machine (SVM) algorithms. The data were collected through a comment scraping technique on related content.The research stages include text preprocessing, sentiment labeling into positive, negative, and neutral categories, and feature extraction using TF-IDF. The classification process employs Naive Bayes and Support Vector Machine algorithms, which are then evaluated based on accuracy, precision, recall, and F1-score. The results of this study indicate that both methods are capable of classifying sentiment effectively. However, the Support Vector Machine method is superior to the Naive Bayes method with an accuracy rate of 99.57% compared to 94.30%. This study is expected to help the government understand public responses to the planned policy of the used mobile phone name transfer system.

The spread of hoax news in digital media is a serious problem because it can affect public opinion and social stability. This study aims to classify hoax news using the Support Vector Machine (SVM) algorithm. The dataset used is a hoax clarification dataset from the Ministry of Communication and Digital (Komdigi) of the Republic of Indonesia, totaling 1,872 data. The research process includes data collection, text pre-processing, feature extraction using TF-IDF, and classification using the SVM algorithm. Implementation was carried out using Google Colaboratory (Google Colab). Test results show that the SVM algorithm is able to provide good performance in classifying hoax news based on its topic with satisfactory accuracy, precision, recall, and F1-score values.

The increasing use of Learning Management Systems (LMS) in higher education generates large amounts of student activity data that have the potential to provide deeper insights into learning processes. However, in practice, these data are still rarely analyzed systematically to understand variations in students’ learning activity patterns, limiting their practical use in supporting teaching and learning. This study aims to explore students’ learning activity patterns in an LMS using a clustering approach based on activity data.This research utilizes the publicly available Open University Learning Analytics Dataset (OULAD), focusing on a single course and a single academic term. LMS activity data were processed through data cleaning and feature extraction, followed by student clustering using the K-Means algorithm. The quality of the clustering results was evaluated using the Silhouette Score, and visual analysis was applied to support the interpretation of the results.The results indicate that students’ learning activities can be grouped into two main patterns, namely a group of students with high learning activity and a group with lower or moderate activity levels. These findings highlight the existence of heterogeneous learning behaviors among students, even within the same learning context.The identified learning activity patterns provide an initial foundation for utilizing LMS data to monitor student engagement and to support the development of more responsive, data-driven learning approaches in higher education.

Savings and Loan Cooperatives (KSP) play a vital role in expanding community access to capital, especially within the informal sector. Nevertheless, non-performing loans remain a persistent challenge that can threaten liquidity and long-term institutional sustainability. KSP CU Mera Ndi Ate faces similar issues, which are assumed to stem not only from administrative weaknesses but also from members’ perceptions and behavioral factors. This research aims to examine the potential causes of non-performing loans through text-based sentiment analysis using an unsupervised learning approach. A quantitative method with a data mining framework was applied. Data were gathered through interviews, observations, documentation, and 200 customer opinion texts processed using the Orange Data Mining application. The analytical stages included preprocessing, corpus development, feature extraction, sentiment clustering, and visualization. Because the dataset lacked predefined labels, unsupervised learning was used to identify naturally emerging sentiment patterns. Findings reveal a predominance of critical sentiments related to credit assessment procedures and service quality. The highest sentiment score (75) concerned insufficient creditworthiness evaluation, followed by concerns about service efficiency (66.6667). These insights suggest that improving assessment accuracy and service quality may help reduce non-performing loans.

Indonesia, as a country with the highest seismicity in the world, requires an accurate earthquake prediction system through the use of the BMKG earthquake catalogue. This research aims to implement ETL-based data pipeline engineering to process 92,887 earthquake catalog entries for the 2008-2023 period into ready-to-use daily time series for the LSTM seismicity forecasting model. The ETL process includes raw data extraction, cleaning of 97% missing values columns on focal mechanism parameters, datetime conversion, daily resampling producing 5,200 entries with earthquake count, total magnitude, and average magnitude features, as well as Min-Max Scaler normalization for LSTM compatibility. The dataset was processed using Google Colab with a stacked LSTM architecture of two layers of 50 and 25 units, dropout 0.2, Adam optimizer, and a sequence window of 30 days to predict the daily earthquake count. The model trained for 100 epochs shows the ability to capture stable seismic activity trends with a consistent decrease in MSE loss, although it shows deviations in extreme spikes due to aftershock sequences. The ETL pipeline proved crucial in ensuring temporal consistency, 100% data completeness, and relevant physics representation, resulting in a reproducible end-to-end framework for disaster mitigation.

This study explores the integration of deep learning based approaches in real time video content analysis for intelligent human computer interaction (HCI) in multimedia systems. Traditional video analysis techniques, such as rule-based methods and offline processing, struggle with real time performance and adaptability to complex video data. In contrast, the deep learning model used in this research, particularly Convolutional Neural Networks (CNNs), provides high accuracy in object detection, feature extraction, and real time processing. The integration of CNNs with interactive visualization modules enables dynamic adjustments to video content based on user interactions, ensuring a seamless and engaging user experience. The system was benchmarked in terms of its processing speed, accuracy, and responsiveness, showing significant improvements over traditional approaches in real time video analysis. Moreover, the study demonstrates that combining deep learning with real time visualization enhances the efficiency of interactive multimedia applications, making it suitable for dynamic environments such as surveillance, security monitoring, and interactive media. Despite the system's strong performance, challenges such as computational demands in high-resolution video processing were identified, highlighting the need for further optimization. Future work will focus on optimizing the system for different hardware platforms, incorporating multimodal inputs, and refining deep learning models to address computational bottlenecks. This research contributes to advancing HCI by providing insights into the integration of deep learning for real time video content analysis, which is pivotal for enhancing the interactivity and adaptability of intelligent multimedia systems.

This study develops a computational-based system to identify individual potential through the analysis of signature patterns using Artificial Neural Networks (ANN) and the Backpropagation algorithm. The research aims to explore and examine the effectiveness of applying ANN in recognizing and identifying signature patterns that are assumed to be related to an individual’s potential. In the data processing stage, Principal Component Analysis (PCA) is employed as a dimensionality reduction and feature extraction technique to optimally obtain the main characteristics of signature images. The system performance evaluation is conducted using a total of 80 signature images, consisting of 60 training data and 20 testing data. This study analyzes two network architecture configurations, namely a model with one hidden layer and a model with two hidden layers. The experimental results show that both network configurations achieve the same accuracy level of 92.5%. These findings indicate that the use of Artificial Neural Networks with the Backpropagation algorithm is effective in producing high accuracy in the signature pattern recognition process. Furthermore, the developed system has broad potential applications in the field of personal identification, such as employee evaluation, selection systems, and other applications across various organizational and industrial sectors.