Claim Missing Document
Check
Articles

Found 4 Documents
Search
Journal : Journal of Applied Data Sciences

Improved Hybrid Machine and Deep Learning Model for Optimization of Smart Egg Incubator Febriani, Anita; Wahyuni, Refni; Mardeni, Mardeni; Irawan, Yuda; Melyanti, Rika
Journal of Applied Data Sciences Vol 5, No 3: SEPTEMBER 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v5i3.304

Abstract

This research develops a Smart Egg Incubator that integrates IoT technology, fuzzy logic, and the YOLOv9-S Deep Learning model to enhance the efficiency and accuracy of hatching chicken eggs. The system automatically regulates temperature and humidity, maintaining temperature between 34.3°C and 39.5°C and humidity between 57% and 68% with a fuzzy logic success rate of 90%. The YOLOv9-S model enables realtime chick detection and classification with mAP50 of 93.7% and mAP50:95 of 71.3%. Efficiency improvements are measured through the success rate of fuzzy logic and improved detection and classification accuracy. This research also uses CNN for high-accuracy object classification, with model optimization performed using SGD to accelerate convergence and improve accuracy. The results indicate significant potential in improving the egg hatching process. The high accuracy and robustness of the YOLOv9-S model enhance real-time monitoring and decision-making in hatcheries, leading to higher hatching success rates, reduced chick mortality, and increased operational efficiency. Future designs can leverage these technologies to create more intelligent, automated systems requiring minimal human intervention, enhancing productivity and scalability. Additionally, IoT and deep learning integration can extend to other poultry farming areas, such as broiler production and disease monitoring, providing a comprehensive approach to farm management. Future research could focus on integrating the YOLOv10 model for even higher accuracy and efficiency, exploring diverse data augmentation techniques, optimizing fuzzy logic algorithms, and integrating additional sensors like CO2 and advanced humidity sensors to improve environmental regulation. These advancements would benefit not only smart incubator applications but also broader poultry farming areas.
A Comprehensive Stacking Ensemble Approach for Stress Level Classification in Higher Education Fonda, Hendry; Irawan, Yuda; Melyanti, Rika; Wahyuni, Refni; Muhaimin, Abdi
Journal of Applied Data Sciences Vol 5, No 4: DECEMBER 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v5i4.388

Abstract

This research focuses on developing a comprehensive ensemble stacking model for the classification of student stress levels in higher education environments, specifically at Hang Tuah University Pekanbaru. Using a physiological dataset that includes parameters such as SPO2, heart rate, body temperature, systolic, and diastolic pressure, this research categorizes the condition of college students into four main categories: anxious, calm, tense, and relaxed. The data taken from public health centers in the period 2021 to 2024 was processed using the SMOTE technique to overcome data imbalance and K-Fold Cross Validation for model validation. In model development, a combination of basic algorithms such as SVM, Logistic Regression, Multilayer Perceptron, and Random Forest is used which is enhanced by boosting techniques through ADABoost, and XGBoost as a meta model. The test results show that the proposed stacking model is able to achieve 95% accuracy, with an AUC of 0.95, which indicates excellent performance in classification. The model not only excels in detecting more extreme stress conditions such as anxiety, but also shows reliable ability in classifying more difficult to distinguish conditions such as tense and relaxed. The conclusion of this study shows that the applied stacking ensemble approach significantly improves prediction accuracy and stability compared to traditional models. For future research, it is recommended to explore the use of deep learning-based meta-models such as LSTM and BiLSTM as well as rotation techniques in stacking to improve model performance and flexibility. The findings are expected to contribute significantly to the development of more sophisticated and effective stress detection models.
Optimization of Machine Learning Models for Risk Prediction of DHF Spread to Support Management Strategies in Urban Areas Devis, Yesica; Muhamadiah, Muhamadiah; Yulanda, Yulanda; Irawan, Yuda; Wahyuni, Refni
Journal of Applied Data Sciences Vol 6, No 4: December 2025
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v6i4.898

Abstract

Dengue fever is an endemic disease that poses a serious threat to public health in tropical regions such as Indonesia. Efforts to control this disease require a data-based approach that is able to accurately predict the level of risk so that interventions can be targeted. This study aims to develop a predictive model of DHF risk using ensemble stacking method optimized with Optuna algorithm and integrated into an interactive dashboard based on Streamlit. The dataset used includes environmental, climate, and socio-demographic indicators from 2015 to 2024 with a total of 1,440 data entries. The preprocessing process includes normalization with Standard Scaler, feature selection using LASSO, and label data balancing with the SMOTE method. Model validation was performed using 10-Fold Cross Validation to ensure model generalization to new data. The stacking model is built with three basic algorithms, namely SVM, KNN, and Random Forest, which are combined using Logistic Regression as a meta-learner. The evaluation results show that the model is able to achieve an average accuracy of 97.57%, with high precision, recall, and f1-score values in all three prediction classes (low, medium, high). The ROC-AUC for each class also showed near-perfect performance. The implementation of the model in the Streamlit dashboard allows non-technical users such as health center or health office staff to perform regional risk prediction and obtain data-driven intervention recommendations automatically. This research not only contributes to the development of predictive technology, but also strengthens evidence-based health promotion practices in urban areas. Further research is recommended to integrate IoT-based real-time data and expand the scope of application areas.
Multimodal Deep Learning and IoT Sensor Fusion for Real-Time Beef Freshness Detection Kurniawan, Bambang; Wahyuni, Refni; Yulanda, Yulanda; Irawan, Yuda; Habib Yuhandri, Muhammad
Journal of Applied Data Sciences Vol 6, No 4: December 2025
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v6i4.977

Abstract

Beef freshness quality is one of the important indicators in ensuring food safety and suitability. However, conventional methods such as manual visual inspection and laboratory testing cannot be widely applied in real-time and mass scale. To overcome these challenges, this study proposes a meat freshness detection system based on a multimodal approach that combines visual imagery and gas sensor data in a single IoT-based framework. This system is designed by utilizing the YOLOv11 architecture that has been optimized using the Adam optimizer. The dataset consisted of 540 original beef images, expanded into 1,296 images after augmentation. The model is trained on these augmented images and is able to achieve detection performance with a mAP@0.5 value of 99.4% and mAP@0.5:0.95 of 95.7%. As a further improvement, the cropped image features from the YOLOv11 model are processed through a combination of the ViT model and CNN to classify the level of meat freshness into three classes: Fresh, Medium, and Rotten with an accuracy of 99%. On the other hand, chemical data was obtained from the MQ136 and MQ137 gas sensors to detect H₂S and NH₃ levels which are indicators of meat spoilage. Data from visual and chemical data were then combined through a multimodal fusion method and classified using the Random Forest algorithm, producing a final prediction of Fit for Consumption, Need to Check, and Not Fit for Consumption. This multimodal model achieved a classification accuracy of 98% with a ROC-AUC score approaching 1.00 across all classes. While the proposed system achieved very high accuracy, further validation across diverse real-world environments is recommended to establish its generalizability.