Claim Missing Document
Check
Articles

Found 13 Documents
Search

Type Deep Learning Model for Multi-Label Waste Classification in Canal Environments: A Comparative Study with CNN Architectures Najirah Umar; Billy Eden William Asrul; Yuyun Yuyun
Journal of Applied Data Sciences Vol 7, No 1: January 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i1.1066

Abstract

The escalating environmental degradation caused by waste underscores the necessity of developing intelligent and sustainable management systems. This study introduces a deep learning–based framework with proposed a modified ConvNeXt architecture enhanced by a two-layer non-linear MLP classification, specifically designed for multi-object waste classification in canal environments. Specifically, ConvNeXt-CNN is introduced as the primary backbone for extracting visual features from waste images. Then, a modified Multi-Layer Perceptron (MLP) is employed to transform these features into multi-label predictions. To optimize the model’s generalization capability in recognizing the complexity of waste images, a hybrid data augmentation technique combining SMOTE and MixUp was applied during training. The proposed approach was then compared with ten fine-tuned Convolutional Neural Network (CNN) architectures, ResNet18, ResNet50, VGG16, VGG19, DenseNet121, MobileNet_v2, and EfficientNet (B0, B1, B2, and B3), and evaluated using accuracy, precision, recall, and F1-score metrics. The experimental dataset comprises 855 waste images containing a total of 2,662 annotated objects across 18 categories, including Bamboo, Beverage Carton, Cardboard, Fabric, Glass Bottle, Inorganic Waste, Kite, Leaf, Metal, Organic Waste, Paper, Plastic, Plastic Bottle, Plastic Cup, Residual Waste, Rubber, Small E-waste, Styrofoam, and Wood. The results show that the fine-tuned ConvNeXt achieved the best performance with an F1-score of 0.99, surpassing DenseNet121 (0.95), ResNet18 (0.91), and VGG16 (0.94). The ConvNeXt model demonstrated its robust capability by achieving consistently high identification scores across majority 18 waste categories. When it came to training efficiency, the fine-tuned MobileNetV2 model proved to be the top performer, outclassing ten other pretrained models, with a training time of 13.35s per epoch.  Results exhibit that finetuned ConvNext outperforms in terms of accuracy, recall, precision, and F1-score. In conclusion, Integrating ConvNeXt and MLP for multi-object waste classification effectively supports intelligent waste management, enabling practical real-world deployment in smart bins, Material Recovery Facilities, and IoT-integrated urban waste systems.
Type Deep Learning Model for Multi-Label Waste Classification in Canal Environments: A Comparative Study with CNN Architectures Najirah Umar; Billy Eden William Asrul; Yuyun Yuyun
Journal of Applied Data Sciences Vol 7, No 1: January 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i1.1066

Abstract

The escalating environmental degradation caused by waste underscores the necessity of developing intelligent and sustainable management systems. This study introduces a deep learning–based framework with proposed a modified ConvNeXt architecture enhanced by a two-layer non-linear MLP classification, specifically designed for multi-object waste classification in canal environments. Specifically, ConvNeXt-CNN is introduced as the primary backbone for extracting visual features from waste images. Then, a modified Multi-Layer Perceptron (MLP) is employed to transform these features into multi-label predictions. To optimize the model’s generalization capability in recognizing the complexity of waste images, a hybrid data augmentation technique combining SMOTE and MixUp was applied during training. The proposed approach was then compared with ten fine-tuned Convolutional Neural Network (CNN) architectures, ResNet18, ResNet50, VGG16, VGG19, DenseNet121, MobileNet_v2, and EfficientNet (B0, B1, B2, and B3), and evaluated using accuracy, precision, recall, and F1-score metrics. The experimental dataset comprises 855 waste images containing a total of 2,662 annotated objects across 18 categories, including Bamboo, Beverage Carton, Cardboard, Fabric, Glass Bottle, Inorganic Waste, Kite, Leaf, Metal, Organic Waste, Paper, Plastic, Plastic Bottle, Plastic Cup, Residual Waste, Rubber, Small E-waste, Styrofoam, and Wood. The results show that the fine-tuned ConvNeXt achieved the best performance with an F1-score of 0.99, surpassing DenseNet121 (0.95), ResNet18 (0.91), and VGG16 (0.94). The ConvNeXt model demonstrated its robust capability by achieving consistently high identification scores across majority 18 waste categories. When it came to training efficiency, the fine-tuned MobileNetV2 model proved to be the top performer, outclassing ten other pretrained models, with a training time of 13.35s per epoch.  Results exhibit that finetuned ConvNext outperforms in terms of accuracy, recall, precision, and F1-score. In conclusion, Integrating ConvNeXt and MLP for multi-object waste classification effectively supports intelligent waste management, enabling practical real-world deployment in smart bins, Material Recovery Facilities, and IoT-integrated urban waste systems.
Coordinate Identity Features for Unsupervised Detection of Attendance Location Fraud Nur Trismi Auliah; Wardi Wardi; Hazriani Hazriani; Syafruddin Syarif; Yuyun Yuyun; Imran Taufik
Journal of System and Computer Engineering Vol 7 No 4 (2026): JSCE: October 2026
Publisher : Universitas Pancasakti

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.61628/jsce.v7i4.2972

Abstract

Web-based attendance systems cannot read the mock location flag of a user device, thus, the coordinates produced by spoofing applications pass geofence validation unnoticed. This study asks whether detection can be moved from the device to the historical pattern, and which unsupervised detector suits the data that result. Eleven composite spatio-temporal-behavioural features were engineered from the attendance log of a corporate human resource information system in Indonesia, three of which measured how often an employee repeated the same coordinate at six-decimal precision. Isolation Forest, a one-class support vector machine, and a long short-term memory autoencoder were compared under twelve-fold walk-forward cross-validation on 249,818 records from 212 employees, evaluated against pseudo-labels derived from seven documented attendance rules. The stored models were then applied to 18,711 later records, and the leading indicator was confirmed with ten employees through a blinded matched case-control interview. The one-class support vector machine led on every accuracy metric, reaching a precision-recall area of 0.5367 and a Matthews correlation coefficient of 0.2541, and its ranking held across three calibration scenarios. Removing the coordinate repetition features cost 74.2 percent of the discriminative margin, while the field confirmation separated flagged from comparison employees completely.