GAGregorius Airlangga

Gregorius Airlangga

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Gregorius Airlangga has a total of 36 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Systematic Diagnosis and Benchmarking of Memory Systems in Autonomous AI Research Agents: A Low-Resource Framework

    by Nur Arifin Akbar, Rahool Dembani, Gregorius Airlangga, Ripto Mukti Wibowo, Biagio Lenzitti, Domenico Tegolo


  • Open Access

    Evaluating Hybrid and Sampling-Based Path Planners for Unmanned Surface Vehicle Logistics in Dynamic Inland Waterways

    by Ronald Sukwadi, Gregorius Airlangga, Widodo Widjaja Basuki, Lai Ferry Sugianto, Emanuel Jando, Dimas Catur Winson

    Autonomous surface vehicles (ASVs), particularly unmanned surface vehicles (USVs), play an increasingly important role in supporting logistics across Indonesia’s archipelagic and riverine regions, where shallow depths, strong currents, and debris complicate navigation. The research contribution is the development of a unified USV logistics simulation framework that enables systematic benchmarking of multiple path-planning strategies under realistic inland-waterway conditions. The framework integrates hydrodynamic environmental modeling, stochastic mission generation, and multi-planner evaluation within a single experimental loop. Nine algorithms were implemented, including classical planners (A*, D* Lite, GA, APF, Reactive Steering), sampling-based methods (RRT, RRT*), and hybrid variants (HARRT*, AHA-RRT*) that employ current-aware travel-time costs and adaptive replanning. Experiments conducted on a 200 × 200 m simulated domain with a fleet of three USVs under a 10-minute deadline revealed that sampling-based planners, particularly RRT*, achieved complete mission success with zero collisions, efficient energy use, and high obstacle clearance. The adaptive hybrid AHA-RRT* also performed robustly, showing superior adaptability albeit with longer paths due to frequent replans. In contrast, classical and reactive methods failed to meet time and safety constraints. These findings demonstrate that sampling-based and hybrid planners offer the most effective balance of safety, efficiency, and adaptability for autonomous inland-waterway logistics, providing a validated framework to guide future USV deployment in dynamic archipelagic environments.

  • Open Access

    Hybrid CNN-LSTM Model with Custom Activation and Loss Functions for Predicting Fan Actuator States in Smart Greenhouses

    by Gregorius Airlangga, Julius Bata, Oskar Ika Adi Nugroho, Boby Hartanto Pramudita Lim

    Smart greenhouses rely on precise environmental control to optimize crop yields and resource efficiency. In this study, we propose a novel hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture to predict fan actuator states based on environmental data. The hybrid model integrates CNNs for spatial feature extraction and LSTMs for temporal dependency modeling, enhanced by a custom activation function and loss function tailored for the problem’s characteristics. The model was trained and evaluated on a comprehensive dataset containing 37,923 samples with 13 environmental features, collected from a smart greenhouse. Experimental results demonstrate the superior performance of the hybrid CNN-LSTM model, achieving an accuracy of 0.9992, precision of 0.9989, recall of 0.9996, and an F1 score of 0.9992, significantly outperforming traditional machine learning methods such as Random Forest and Gradient Boosting, as well as standalone CNN and LSTM architectures. The high recall underscores the model’s reliability in identifying positive actuator states, critical for greenhouse management. This study highlights the importance of hybrid architectures in handling complex spatiotemporal data, offering potential applications beyond greenhouses, such as healthcare monitoring and predictive maintenance. Despite the model’s strengths, limitations include computational complexity and limited interpretability, necessitating future work on optimization and explainability. These findings establish a foundation for integrating deep learning into smart agricultural systems, advancing the automation and efficiency of environmental control mechanisms.


  • Open Access

    A Comparative Analysis of Deep Learning Architectures for Obesity Classification Using Structured Data

    by Gregorius Airlangga

    Obesity is a significant global health concern, necessitating accurate and efficient diagnostic tools to classify individuals based on obesity levels. This study investigates the performance of five deep learning architectures: Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional LSTM (BiLSTM) in classifying obesity levels using structured data. The dataset comprises clinical, demographic, and lifestyle features, and is preprocessed through normalization, label encoding, and Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. Each model was evaluated using accuracy, precision, recall, and F1-score metrics under stratified 10-fold cross-validation. The results indicate that MLP achieved the highest performance across all metrics, with an accuracy of 99.05%, followed closely by CNN at 98.77%. Sequential models, including LSTM, GRU, and BiLSTM, exhibited comparatively lower performance, achieving accuracies of 83.80%, 86.59%, and 86.78%, respectively. The superior performance of MLP and CNN underscores their suitability for structured datasets with static features, while the sequential models struggled due to the lack of temporal dependencies in the data. This study highlights the importance of aligning model architecture with dataset characteristics for optimal performance. The findings suggest that MLP and CNN are effective choices for obesity classification tasks, providing robust and computationally efficient solutions. Future work could explore hybrid models and incorporate temporal features to enhance the performance of sequential architecture.


  • Open Access

    A Comparative Analysis of Machine Learning Models for Obesity Prediction

    by Gregorius Airlangga

    Obesity is a global health challenge with significant implications for public health systems and individual well-being. Predictive modeling using machine learning (ML) offers a powerful approach to identify individuals at risk of obesity and inform early intervention strategies. This study evaluates the performance of ten ML models, including Logistic Regression, Support Vector Machines, Decision Trees, K-Nearest Neighbors, Naive Bayes, Random Forest, Gradient Boosting, AdaBoost, XGBoost, and LightGBM, in predicting obesity using a publicly available dataset. A rigorous preprocessing pipeline, incorporating missing value handling, categorical encoding, normalization, and outlier detection, was applied to ensure data quality and compatibility with ML algorithms. Performance metrics such as accuracy, precision, recall, and F1-score were evaluated using 10-fold stratified cross-validation. Among the models, LightGBM demonstrated the highest test accuracy (99.19%) and F1-score (99.20%), outperforming Gradient Boosting and Random Forest, which also showed competitive results. The study highlights the superior predictive capabilities of ensemble methods while underscoring the trade-offs between model complexity and interpretability. Logistic Regression provided a strong baseline, demonstrating the importance of preprocessing, but was outperformed by advanced ensemble techniques. This research contributes to the growing field of ML-driven healthcare solutions, offering valuable insights into the strengths and limitations of various predictive models. The findings support the integration of advanced ML techniques in public health systems and pave the way for future research on hybrid and explainable models for obesity prediction and management.


  • Open Access

    Machine Learning-Based Obesity Classification: A Comparative Study Using Self-Reported Survey Data and Ensemble Learning Models

    by Gregorius Airlangga

    Obesity has become one of the most pressing global health challenges of the 21st century, with its prevalence increasing at an alarming rate. Obesity is a major global health concern, contributing to an increased risk of cardiovascular disease, diabetes, and other metabolic disorders. Traditional assessment methods, such as BMI-based classification, often fail to incorporate lifestyle and behavioral factors, limiting their predictive capabilities. This study explores the use of machine learning for obesity classification based on self-reported survey data collected from individuals in Mexico, Peru, and Colombia. The dataset comprises 2111 instances with 17 attributes, covering demographic characteristics, eating habits, and physical activity levels. Eight machine learning models, including Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machine (SVM), Decision Tree, K-Nearest Neighbors, Naïve Bayes, and AdaBoost, were evaluated using 10-fold cross-validation. Results indicate that Gradient Boosting achieved the highest accuracy of 96.49%, followed by Random Forest and SVM, demonstrating the effectiveness of ensemble learning techniques in capturing complex feature interactions. In contrast, Naïve Bayes and AdaBoost exhibited the lowest classification performance due to their strong assumptions about feature independence and sensitivity to noisy data. The findings highlight the potential of machine learning in obesity classification and underscore the need for advanced predictive models to enhance public health monitoring and intervention strategies.


  • Open Access

    Hybrid Machine Learning Approaches for Atmospheric CO₂ Prediction: Evaluating Regression and Ensemble Models with Advanced Feature Engineering

    by Gregorius Airlangga

    The accurate prediction of atmospheric CO₂ concentrations is essential for understanding climate change dynamics and developing effective environmental policies. This study evaluates the predictive capabilities of various machine learning models, including ensemble-based regressors such as Random Forest, Gradient Boosting, and XGBoost, alongside traditional regression models such as Support Vector Regression (SVR), Ridge, and Lasso regression. The dataset, derived from meteorological observations, was preprocessed using multiple feature scaling techniques, including StandardScaler, MinMaxScaler, and RobustScaler, followed by feature engineering techniques such as polynomial transformation and Principal Component Analysis (PCA) to enhance predictive accuracy. Model performance was assessed using the coefficient of determination (R²) and cross-validation techniques. The results indicate that tree-based models, including Random Forest and XGBoost, struggled to generalize well, exhibiting negative R² values due to overfitting and an inability to capture the temporal dependencies in CO₂ variations. SVR emerged as the best-performing model, though its predictive power remained limited. Computational complexity analysis revealed that tree-based methods incurred high processing costs, while linear models such as Ridge and Lasso demonstrated lower complexity but failed to capture non-linear dependencies. The study highlights the challenges of CO₂ prediction using conventional machine learning techniques and underscores the need for advanced deep learning approaches, such as hybrid Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models, to better capture spatial and temporal dependencies. Future research should explore integrating external environmental factors and leveraging deep learning architectures to improve predictive performance.

  • Open Access

    Comparative Study of Machine Learning Models for Temperature Prediction: Analyzing Accuracy, Stability, and Generalization

    by Gregorius Airlangga

    Accurate temperature prediction is crucial for climate monitoring, energy management, and disaster preparedness. This study provides a comparative analysis of various machine learning models, including Random Forest, Gradient Boosting, Histogram-Based Gradient Boosting, XGBoost, Support Vector Regression (SVR), Ridge Regression, and Lasso Regression, to evaluate their predictive accuracy, stability, and generalization capability. The models are assessed using five-fold cross-validation, with the R² metric as the primary evaluation criterion. The results indicate that Random Forest achieves the highest accuracy, with an R² mean of 0.999994, demonstrating its strong ability to model temperature variations. Ridge Regression unexpectedly performs at a similar level, suggesting that the dataset contains strong linear dependencies. Gradient Boosting, Histogram-Based Gradient Boosting, and XGBoost also achieve high accuracy, confirming their effectiveness in capturing complex relationships between meteorological parameters. SVR, while effective, exhibits higher variance, indicating that it may require further tuning for improved consistency. Lasso Regression, with an R² mean of 0.9783, shows the lowest accuracy, confirming that linear models are less suitable for complex meteorological predictions. These findings highlight the superiority of ensemble-based methods in temperature forecasting, reinforcing their stability and adaptability. Future research should explore hybrid models that integrate ensemble techniques with feature engineering optimizations to further enhance predictive performance. This study contributes to the ongoing development of machine learning applications in meteorology, offering insights into model selection for climate-related forecasting tasks.

  • Open Access

    Enhancing UAV Communication Security: Multi-Label Anomaly Detection Using Machine Learning in Imbalanced Data Environments

    by Gregorius Airlangga, Denny Jean Cross Sihombing, Oskar Ika Adi Nugroho

    Unmanned Aerial Vehicle (UAV) communication networks are increasingly vulnerable to cyber threats, including spoofing, jamming, malware, and distributed denial-of-service (DDoS) attacks. Effective anomaly detection is crucial to maintaining network integrity and operational security. This study evaluates multiple machines learning models, including Support Vector Machines, Logistic Regression, XGBoost, Gradient Boosting, and Random Forest, to detect anomalies in UAV communication networks. A real-world dataset containing 44,016 instances of network telemetry and security indicators was utilized, with each instance labeled for multiple potential anomalies. Experimental results reveal a significant class imbalance, where models achieve high accuracy (92%) but fail to detect minority class anomalies, yielding near-zero recall scores for critical cyber threats. The study highlights the limitations of traditional classifiers in imbalanced multi-label classification tasks and emphasizes the need for advanced techniques such as Synthetic Minority Over-sampling (SMOTE), cost-sensitive learning, and deep learning-based anomaly detection. The findings suggest that conventional machine learning approaches alone are insufficient for reliable anomaly detection in UAV networks, necessitating hybrid solutions that integrate multiple detection paradigms. Future work should explore adaptive ensemble learning methods and deep anomaly detection frameworks to improve recall and precision for rare cybersecurity threats.

  • Open Access

    Detection of DDoS Attacks in UAV Communication Networks Using Machine Learning Models

    by Gregorius Airlangga

    The increasing adoption of unmanned aerial vehicle (UAV) communication networks has introduced new cybersecurity challenges, particularly in detecting and mitigating distributed denial-of-service (DDoS) attacks. This study evaluates the effectiveness of multiple machine learning models, including Random Forest, Gradient Boosting, XGBoost, Logistic Regression, and Support Vector Machine (SVM), for DDoS attack detection in UAV networks. The dataset, derived from a simulated UAV communication network, incorporates key network parameters such as signal strength, packet loss rate, round-trip time, and base station load. Data preprocessing steps, including feature selection, normalization, and synthetic minority over-sampling (SMOTE), were applied to enhance model performance. Among the evaluated models, Random Forest demonstrated the highest classification accuracy with an F1-score of 0.839 and an AUC score of 0.912, outperforming other models in precision-recall trade-offs. Gradient Boosting and XGBoost exhibited moderate classification ability, whereas Logistic Regression and SVM struggled with capturing complex network patterns. The results highlight the effectiveness of ensemble learning in intrusion detection for UAV networks. This study provides valuable insights into optimizing machine learning-based intrusion detection systems and paves the way for further advancements in UAV cybersecurity. Future work will focus on integrating additional feature engineering techniques and validating models on real-time network traffic datasets.