Android malware is an application that targets Android devices to steal crucial data, including money or confidential information from Android users. Recent years have seen a surge in research on Android malware, as its types continue to evolve, and cybersecurity requires periodic improvements. This research focuses on detecting Android malware attack patterns using deep learning and convolutional neural network (CNN) models, which classify and detect malware attack patterns on Android devices into two categories: malware and non-malware. This research contributes to understanding how effective the CNN models are by comparing the ratio of data used with several epochs. We effectively use CNN models to detect malware attack patterns. The results show that the deep learning method with the CNN model can manage unstructured data. The research results indicate that the CNN model demonstrates a minimal error rate during evaluation. The comparison of accuracy, precision, recall, F1 Score, and area under the curve (AUC) values demonstrates the recognition of malware attack patterns, reaching an average of 92% accuracy in data testing. This provides a holistic understanding of the model's performance and its practical utility in detecting Android malware.

M.T., Ph.D., IPU., ASEAN Eng.

Prof. Ir.

Deris Stiawan

Detection of android malware with deep learning method using convolutional neural network model

The low level of students’ understanding in learning Natural Sciences, particularly on renewable energy topics, in elementary schools is partly due to the lack of interactive learning media that align with children's learning characteristics. This study aims to analyze the needs of both teachers and students for the development of interactive infographic media grounded in the discovery learning approach. A quantitative descriptive method was employed, using closed-ended questionnaires as the primary data collection instrument. The respondents consisted of 30 students and six classroom teachers from an elementary school in Musi Banyuasin Regency. The questionnaire covered pedagogical, technical, cognitive, aesthetic, and evaluation aspects. The findings revealed that the majority of students struggled to comprehend renewable energy concepts in the absence of visual media. Furthermore, 90% of students expressed a preference for learning materials incorporating images, videos, and animations. All teacher respondents emphasized the need for media that supports active and visual learning tailored to the developmental characteristics of elementary students. These findings underscore the urgent need to develop interactive infographic media based on discovery learning to enhance students’ conceptual understanding. Practically, the results serve as a foundation for designing learning tools that are engaging, age-appropriate, and capable of fostering independent exploration in science education at the elementary level.

Needs Analysis for the Development of Interactive Infographic Media Based on Discovery Learning on Renewable Energy Topics in Elementary Education

FC-PoW: A Machine Learning-Based Blockchain Consensus for Efficient Order Processing in Manufacturing

The proliferation of cyber threats has created a deluge of unstructured cyber threat intelligence (CTI), making manual analysis impractical and over-whelming. Knowledge graphs (KGs) offer a powerful solution for structur-ing this data, but their construction is bottlenecked by the foundational task of named entity recognition (NER). While state-of-the-art NER often relies on massive general-purpose or resource-intensive domain-specific models, this paper explores a novel alternative that leverages a lightweight, code-centric language model, DeepSeek Coder 1.3B. We hypothesize that its unique pre-training on a 2-trillion-token corpus, heavily skewed towards source code (87%), provides a strong inductive bias for recognizing the quasi-syntactic nature of cybersecurity entities. We fine-tune DeepSeek Coder on the CyberNER dataset, a public corpus derived from security blogs and annotated with 10 distinct entity types using the BIO tagging scheme to test our hypothesis. This research evaluates the viability of using code-specialized models as an efficient and effective method for populating cybersecurity knowledge graphs, addressing a critical challenge in automat-ing threat intelligence analysis.

S.Kom.

Dendi Renaldo Permana

S.Kom., M.Kom.

Nurul Afifah

Septiani Kusuma Ningrum

Automating Threat Intelligence Knowledge Graph Construction through Named Entity Recognition with DeepSeek

Mental health, encompassing emotional, psychological, and social well-being, plays a pivotal role in individuals’ academic and personal success. Among undergraduate students, the prevalence of mental health disorders such as anxiety, depression, and stress has escalated in response to academic pressures, lifestyle transitions, and socioeconomic factors. Early detection is vital for mitigating severe repercussions, including diminished academic performance and increased risk of self-harm. Leveraging the Depression, Anxiety, and Stress Scale-42 (DASS-42), this study employs machine learning (ML) methods—specifically Support Vector Machine (SVM) and Random Forest (RF)—to analyze mental health patterns under a two-class scenario (“Good” vs. “Counseling”). A balanced dataset was obtained through the Synthetic Minority Over-Sampling Technique (SMOTE), ensuring fairer representation of minority “Counseling” cases. The results demonstrate that both SVM and RF excel in detecting depression, each achieving near-perfect accuracy, precision, and recall (0.97). This performance suggests that the discriminative power of the features for depression is sufficiently robust to enable clear classification boundaries with minimal misclassification. For anxiety, SVM attains a slightly higher accuracy (0.91) compared to RF (0.90), largely driven by SVM’s superior precision (0.94) when identifying “Good” cases. In stress prediction, SVM again outperforms RF (0.95 vs. 0.93 accuracy), although RF showcases near-perfect precision and recall for “Counseling.” Despite these successes, both models occasionally falter when distinguishing borderline cases, indicating that feature overlap and remaining class imbalance still pose challenges. In conclusion, SVM and RF provide promising avenues for early mental health detection in university settings, highlighting the potential of ML-driven tools to inform targeted interventions. Future research should focus on expanding feature sets (e.g., integrating behavioral or physiological data) and refining sampling strategies to enhance minority-class detection.

S.Kom., M.T., Ph.D.

Muhammad Qurhanul Rizqie

Early Mental Health Detection with Machine Learning: A Practical Approach to Model Development and Implementation

Haystack is Facebook's unique way of managing large amounts of user-generated content like photos. The architecture prioritizes performance, reliability, and scalability to overcome network-attached storage system bottlenecks. Haystack speeds data access and ensures data integrity during hardware failures by using physical and logical volumes. This study examines the architecture of Facebook's Haystack data storage system and its effects on scalability and efficiency in handling large photo data. According to the study, the store, directory, and cache functions work together to reduce input/output (I/O) operations and improve metadata processing, which traditional network-attached storage systems cannot do. Haystack manages massive photo data storage and retrieval, solving network-attached storage (NAS) limitations. It balances throughput and latency by minimizing disk operations and optimizing metadata processing. Each store, directory, and cache contribute to this ecosystem. The Haystack architecture reduces disk operations and metadata processing bottlenecks with distributed caching. A cache allows instant access to frequently requested images and balances read and write operations across the system. We should study advanced storage system architectures based on Facebook's Haystack architecture. This could involve investigating faster metadata processing algorithms, using artificial intelligence (AI) to improve fault detection and repair systems, and assessing the economic impact of distributed caches.

S.Kom., M.T.

Dr. Ir.

Ahmad Heryanto

Haystack-based Facebook’s data storage architecture: store, directory, and cache

Hydroponics is a unique and fascinating farming technique for producing plants and vegetables. Without having to use a large area of land, people can easily apply the technique to produce fresh and hygienic vegetables. However, the technique cannot be used in apartment environment due to the limited sunlight. Thus, this study introduces an innovative hydroponic system, called as hydroponics smart showcase system that can be implemented indoors, even in the presence of minimal sunlight, and can be monitored online by users. The proposed system consists of a net pot of 4-5 hydroponics cups with a diameter of 50 mm, air temperature and humidity sensors, water level sensors, ultraviolet (UV) lights, indicator displays, and DC fans. Experimental results show that the development of innovative hydroponics using smart showcase has succeeded in stabilizing the air in the showcase according to the specified references. Moreover, UV light intensity settings for photosynthesis can be applied remotely with duration of 24 hours.

Innovative smart showcase design for indoors and eco-friendly hydroponics

In scientific research, the identification process of finding various relevant methods is carried out through in-depth analysis of various literatures. This analysis aims to evaluate the contribution of previous research novelties and extract valuable information from existing literature. This study focuses on the development of an abstract text extraction model from scientific literature using a machine learning model. The scientific literature abstract text dataset used in this study was obtained from the Garuda National Aggregator. The extraction models used include Conditional Random Field (CRF), Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (Bi-LSTM) to read and extract abstract text from scientific literature, as well as identify important entities in unstructured text. The results of the study show that the Bi-LSTM model is able to provide good performance in extracting entities from abstract text with Precision, Recall, and F1-Score values with a score of 0.92 which indicates its ability to recognize and classify entities with high accuracy and extract information in unstructured abstract text.

The New Model Text Abstract Extraction with Machine Learning

Perkembangan teknologi digital yang pesat menjadikan perangkat mobile seperti smartphone dan tablet sebagai bagian penting dalam kehidupan sehari-hari. Namun, kemajuan ini juga meningkatkan ancaman terhadap keamanan perangkat, terutama melalui malware. Malware adalah perangkat lunak yang dirancang untuk merusak atau mencuri data, dan sering menyebar melalui aplikasi yang tidak aman. Untuk mengatasi tantangan ini, pengembangan dan implementasi sistem deteksi malware pada perangkat mobile menjadi krusial. Sistem ini bertujuan untuk mengidentifikasi jenis-jenis malware, mendeteksi pola serangan, serta memberikan pemahaman mengenai pentingnya menjaga keamanan informasi dalam penggunaan perangkat mobile dan internet. Manfaat dari kegiatan ini termasuk meningkatkan kemampuan masyarakat dalam menghadapi ancaman malware, memperkuat perlindungan data pribadi, dan mendorong penggunaan internet yang lebih aman. Dengan menyediakan alat yang lebih canggih untuk mendeteksi malware, diharapkan dapat mengurangi insiden keamanan yang disebabkan oleh aplikasi berbahaya dan membangun budaya keamanan siber yang lebih kuat di masyarakat.
KATA KUNCI: Deteksi, Keamanan Siber, Malware, Mobile, Pola Serangan

Adi Hermansyah

Sistem Deteksi Ancaman Malicious Software pada Perangkat Mobile untuk Peningkatan Kesadaran Keamanan Siber

The spread of political hoaxes poses a significant threat to democratic integrity and social stability. This study proposes an efficient and structured machine learning approach to detect political hoax content. The text data undergoes preprocessing, including normalization, removal of punctuation and numbers, tokenization, and stopword elimination. Numerical representation is generated using Term FrequencyInverse Document Frequency (TF-IDF), followed by clustering using the K-Means algorithm to group the text based on feature similarity. The core contribution of this research is the development of an automatic labeling strategy based on political keywords, applied after the clustering stage. Unlike previous studies that rely on pre-labeled datasets or stop at unsupervised clustering, this approach introduces a contextual labeling process to distinguish between political and non-political hoaxes by identifying keyword occurrences within the clustered data. The automatically labeled data is then classified using three classical machine learning algorithms: Logistic Regression, Support Vector Machine (SVM), and Naive Bayes. The models are evaluated to assess their ability to accurately detect political hoaxes. Among them, SVM demonstrates the highest and most consistent classification performance, indicating its suitability for this task. The proposed integrated pipeline proves effective in addressing the challenge of unlabeled data. This study contributes methodologically to the development of adaptive and efficient hoax detection systems that are applicable across various socio-political contexts.

Scientific Publications