Indonesian Journal of Machine Learning and Intelligent Systems
Indonesian Journal of Machine Learning and Intelligent Systems (IJMLIS, Indones. J. Mach. Learn. Intell. Syst., e-ISSN 3164-2756) is a peer-reviewed international journal dedicated to advancing research on theoretical developments and practical implementations in the dynamic fields of machine learning, artificial intelligence, and intelligent systems, published by the Indonesian Artificial Neural Network Society (IdNNS). The journal is committed to maintaining high ethical standards in scholarly publishing and adheres to established guidelines for research integrity and publication ethics. The IJMLIS provides a scientific platform for researchers, academics, policymakers, and industry practitioners to disseminate original findings, innovative methodologies, theoretical advancements, and practical applications of intelligent technologies across various domains. The journal welcomes original research articles and review papers that advance the understanding and practical deployment of intelligent systems across healthcare, finance, cybersecurity, smart cities, and industrial automation. IJMLIS focuses on the development and implementation of machine learning algorithms, data-driven intelligence, and intelligent computing systems that support decision-making, automation, and advanced analytics. The journal encourages interdisciplinary research integrating computer science, data science, artificial intelligence, robotics, and intelligent information systems. The scope of the journal includes, but is not limited to: Learning problems: Clustering, classification, regression, recognition, prediction, and prescription/recommendation; Artificial intelligence and learning methods: Supervised/unsupervised learning, reinforcement learning, ensemble methods, connectionist networks (deep learning), Bayesian networks, and evolutionary-based methods; Intelligent systems structure: intelligent agents, multi-agent systems, expert systems, and cognitive computing; Data and information processing: Data mining, information retrieval, pattern recognition, pattern visualization, image/video processing, voice recognition, and natural language processing (NLP); Applications: Computer vision, AI-powered robotics, intelligent control systems, IoT, cybersecurity and information assurance, financial modeling, bioinformatics, medical imaging, healthcare informatics, game playing, digital governance, and smart cities; System and theoretical analysis: Theoretical frameworks, performance evaluation, algorithm design, and computational complexity analysis; Emerging technologies: Sustainable/green IT, brain-machine interfaces, and human-centered AI
Articles
5 Documents
<![CDATA[Comparative study of NBC, SVM, and CNN performance in sentiment analysis of the millennial farmers program on platform X ]]>
<![CDATA[Adamu Abu Bakar Ibrahim]]>;
<![CDATA[Junaedi Rahmat]]>;
<![CDATA[Deshinta Arrova Dewi]]>
<![CDATA[ Indonesian Journal of Machine Learning and Intelligent Systems Vol. 1 No. 1 (2026) ]]>
Publisher : <![CDATA[Indonesian Artificial Neural Network Society ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
<![CDATA[Public perception of government programs is often expressed on social media, making sentiment analysis a vital tool for understanding public opinion. This research presents a comparative study of three machine learning algorithms, Naive Bayes Classifier (NBC), Support Vector Machine (SVM) and Convolutional Neural Network (CNN), applied to sentiment analysis on the Millennial Farmers Program 2024 from Platform X. The study aims to measure the performance of each algorithm in terms of accuracy, execution time, and memory efficiency by adopted a quantitative approach with a comparative experimental design. Sentiment data were collected from platform X (Twitter) between October 2024 and January 2025, with a total of 5,177 raw tweets acquired using the keyword "petani_milenial". After data preprocessing steps, including cleaning, case folding, tokenization, stopword removal, and stemming, the data's sentiment (positive, negative, neutral) was automatically labeled using the IndoBERT model. The sentiment analysis results revealed a dominance of neutral sentiment (832 tweets), followed by negative (331 tweets) and positive (197 tweets). Model performance evaluation showed that SVM achieved the highest overall accuracy at 79%, demonstrating superior capability in classifying neutral sentiment. CNN, with an accuracy of 74.26%, stood out in recall for positive sentiment (73%), making it an effective choice for comprehensive identification of positive sentiment. Meanwhile, NBC, with 78% accuracy, proved to be the most efficient in terms of computation time and memory usage. The study concludes that the optimal model selection greatly depends on the specific use-case priority, whether overall accuracy, positive sentiment identification, or computational efficiency. The results show that CNN outperforms NBC and SVM in accuracy, while NBC is the fastest and consumes the least memory. These findings can help inform future research and implementation in large-scale sentiment analysis applications.]]>
<![CDATA[Improving spam detection techniques using machine learning algorithms ]]>
<![CDATA[Abdullah Said Alkalbani]]>;
<![CDATA[Iin Lesmana Putra]]>;
<![CDATA[Dini Oktarina Dwi Handayani]]>
<![CDATA[ Indonesian Journal of Machine Learning and Intelligent Systems Vol. 1 No. 1 (2026) ]]>
Publisher : <![CDATA[Indonesian Artificial Neural Network Society ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
<![CDATA[Spam detection has become a critical challenge in maintaining the integrity and security of digital communication systems. This study focuses on enhancing spam detection precision by applying the XGBoost algorithm, compared to other machine learning models such as Support Vector Machines (SVM) and Random Forests. Traditional rule-based methods, while straightforward, need to address the increasing sophistication of spam tactics, leading to high rates of false positives and negatives. Machine learning, with its data-driven adaptability, provides a powerful alternative for improving spam detection accuracy. This research evaluates the effectiveness of XGBoost in handling large and complex datasets, leveraging its ability to build robust predictive models by combining weak learners. Performance metrics, including accuracy, precision, recall, and F1-score, are used to assess its efficacy. Additionally, a comparative analysis highlights XGBoost's advantages over SVM, known for its precision in high-dimensional spaces, and Random Forest, valued for its resilience against overfitting. The study emphasizes the importance of parameter optimization and feature engineering in maximizing the performance of spam detection models. This research contributes to the development of more reliable and scalable spam detection frameworks by addressing the diverse attributes of spam messages. The findings offer valuable insights for researchers and practitioners in cybersecurity, underscoring the transformative potential of XGBoost in combating the evolving threat of spam.]]>
<![CDATA[Optimizing Random Forest using Genetic Algorithm for static Android malware detection ]]>
<![CDATA[Youssef Zaz]]>;
<![CDATA[Muhammad Taufiq]]>;
<![CDATA[Barroon Isma'eel Ahmad]]>
<![CDATA[ Indonesian Journal of Machine Learning and Intelligent Systems Vol. 1 No. 1 (2026) ]]>
Publisher : <![CDATA[Indonesian Artificial Neural Network Society ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
<![CDATA[The massive and increasingly complex growth of Android malware has led to a heightened threat to mobile device security. One proven effective approach is the use of machine learning, particularly the Random Forest algorithm. However, many previous studies have not fully utilized the potential of this algorithm, as they rely on default configurations without tuning. This study implements the Genetic Algorithm (GA) method to perform hyperparameter tuning on Random Forest, using the Drebin-215 dataset, which consists of 15,036 APKs and 215 static features. The study compares three configurations: Default and GA Tuning. Evaluation is conducted using 5-fold cross-validation and performance metrics including accuracy, precision, recall, and F1-score. The results show that the GA-Tuned model delivers the best performance, achieving an accuracy of 98.97%. The study also tests the model on 55 real APKs to evaluate its predictive capability in real-world cases.]]>
<![CDATA[TrustEdgeAI: A Lightweight, Calibrated, and Secure Deep Learning Framework for Intrusion Detection in Edge-IoT Environments ]]>
<![CDATA[S Karthika]]>;
<![CDATA[Juliana K. Gnanaselvi]]>;
<![CDATA[Sivakumar N. Sellappan]]>;
<![CDATA[M Gokilavani]]>;
<![CDATA[Hasan Koten]]>
<![CDATA[ Indonesian Journal of Machine Learning and Intelligent Systems Vol. 1 No. 1 (2026) ]]>
Publisher : <![CDATA[Indonesian Artificial Neural Network Society ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
<![CDATA[The fast development of Edge-IoT systems has posed serious security threats because of its distributed and resource-limited characteristics. The conventional intrusion detection systems are normally ineffective in offering real time and accurate security against the changing cyber threats. To overcome this challenge, this paper presents TrustEdgeAI, a lean and safe deep learning system, which is specialized in effective intrusion detection in Edge-IoT systems. The proposed model will combine a hybrid CNN-LSTM to learn spatial and temporal information of network traffic as well as an adaptive pruning mechanism of features to achieve a reduction in the computational complexity. Moreover, a calibration method based on temperature scaling is also included in order to enhance the credibility of prediction probability in order to make reliable decisions. The experiment outcomes prove that TrustEdgeAI has a high detection rate of 97.8 and much less inference latency of 12 ms and resource use. The framework also has low rates of false positive on various types of attacks and can be very applicable in real time and secure edges deployment. ]]>
<![CDATA[Memory-Augmented AI for Autonomous UAV Search and Rescue: Deep Reinforcement Learning in Complex, Partially Observable Environments ]]>
<![CDATA[Hina Javed]]>;
<![CDATA[Hameed Affifa]]>;
<![CDATA[Imran Ahmad Khan]]>
<![CDATA[ Indonesian Journal of Machine Learning and Intelligent Systems Vol. 1 No. 1 (2026) ]]>
Publisher : <![CDATA[Indonesian Artificial Neural Network Society ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
<![CDATA[Unknown, partially observable environments The search of autonomous targets remains one of the basic problems of embodied artificial intelligence. We also present RecurrentPPO-LSTM, an agent of memory-based deep reinforcement learning, which not only attains best-in-class state-of-the-art search efficiency on standard Partially Observable Markov Decision Process (POMDP) benchmarks, but also trains 3.4x faster than other memory-augmented agents. As compared to existing schemes that rely on the bidirectional recurrence (orpruing causality) or complex offline pretraining, our agent can be trained using sparse success signals using unidirectional LSTM to guarantee compact belief states. Our implementation on MiniGrid DoorKey-8x8 has 82.4 percent, which corresponds with higher performance than memoryless Proximal Policy Optimization (65.1 percent) and Deep Q-Networks (58.2 percent). We demonstrate that we have 34 steps-to-target reduction and improved resistance to observation noise (sigma = 0.2), arbitrary dropout sporadically (50 percent), and stochastic transitions (slip probability = 0.3). The transfer learning when trained on photorealistic Habitat Matterport 3D has a 68 percent success with 10,000 fine-tuning steps only. It is relatively compared to five stateof-the-art designs (20252026) to confirm that simple causal memory designs are much better than complex bidirectional designs or modular ones when considering real-time autonomous search. We demonstrate in theory that LSTM augmentation causes minimized anticipated regret of finite-memory POMDPs and we can formally ensure empirical excellence.]]>
