Journal Computer and Technology

A Hybrid YOLOv5-SqueezeNet Framework with Crop-Union Context Verification for Real-Time Motorcycle Rider Identification

2025-12-29T14:01:34+07:00

Motorcycles are the dominant transportation mode in developing nations, yet their high volume exacerbates traffic violations and accidents. Automated identification of motorcycle operators is crucial for Intelligent Transportation Systems (ITS). However, existing advanced vision models, such as the Segment Anything Model (SAM) or YOLOv9, demand substantial computational resources, rendering them impractical for real-time edge deployment. Furthermore, standard object detectors often fail to contextually distinguish active riders from bystanders. To address these limitations, this study proposes a lightweight, context-aware hybrid framework integrating YOLOv5 for rapid spatial detection and SqueezeNet for semantic contextual verification. The core novelty lies in the Crop-Union Context Verification Mechanism, which extracts the spatial union of paired person-motorcycle bounding boxes, expands it proportionally, and classifies the cropped region using SqueezeNet to confirm the presence of an active operator. Additionally, Contrast Limited Adaptive Histogram Equalization (CLAHE) is applied during preprocessing to enhance image quality under varying illumination. Experimental results demonstrate that the hybrid system successfully identifies motorcycle operators with high spatial confidence (YOLOv5: 0.84 for person, 0.86 for motorcycle) and robust contextual validation (SqueezeNet Top-1 prediction: moped, probability 0.247). By utilizing SqueezeNet’s highly efficient Fire Module architecture (~1.2 million parameters), the proposed framework achieves exceptional computational efficiency without sacrificing discriminative power. Ultimately, the disjunctive fusion of spatial and semantic signals ensures system resilience. This lightweight pipeline proves highly viable for real-time, AI-driven traffic surveillance on resource-constrained edge devices, offering a scalable solution for smart city infrastructure and automated law enforcement.

Unlocking Customer Behavior in COD-based E-commerce: An Unsupervised Learning Approach Using Gaussian Mixture Model for Strategic Segmentation

2025-12-29T13:54:14+07:00

Customer segmentation in Cash on Delivery (COD) logistics is inherently challenging due to highly heterogeneous, skewed transaction data and elevated rejection rates, which frequently limit the efficacy of traditional deterministic clustering models. To address this, this study proposes a robust probabilistic segmentation framework utilizing the Gaussian Mixture Model (GMM) to extract actionable insights from complex COD operations. We analyzed 5,147 raw transaction records, meticulously aggregating them into 2,225 unique customer profiles engineered around three critical logistics-centric features: delivery frequency, total COD value, and average COD value. The model’s optimal configuration was rigorously determined using the Bayesian Information Criterion (BIC) to prevent overfitting, and the entire analytical pipeline was seamlessly integrated into an interactive, web-based Decision Support System (DSS). Empirical results demonstrate that a two-cluster solution optimally balances statistical fit and operational interpretability, clearly delineating "High-Value Active Customers" (68%) from "Low-Value Occasional Customers" (32%). Comparative analysis validates GMM’s superiority over baseline algorithms like K-Means, achieving a significantly higher Silhouette Score (0.64 vs. 0.51) and effectively capturing the non-spherical, overlapping data distributions characteristic of COD behaviors. Crucially, the probabilistic soft-clustering capability of GMM uniquely identified a transitional segment of 15.7% "boundary customers," offering logistics managers a nuanced, early-warning lens for preemptive risk mitigation and targeted interventions. Ultimately, this research bridges the critical gap between advanced probabilistic analytics and operational logistics, providing express delivery enterprises with a scalable, data-driven tool for dynamic route optimization, failed-delivery mitigation, and differentiated service personalization.

A Leakage-Aware Ensemble Framework for Imbalanced Tabular Data: Mitigating SMOTE Contamination in Hotel Cancellation Prediction

2026-06-18T13:02:43+07:00

Predictive modeling on large-scale, imbalanced tabular data is frequently compromised by target leakage and improper resampling, leading to inflated performance metrics. While tree-based ensemble methods like Random Forest (RF) and XGBoost are widely deployed, their architectural divergence in handling complex behavioral anomalies under strict class imbalance remains underexplored. This study proposes a leakage-aware ensemble framework to mitigate SMOTE contamination and target leakage in hotel cancellation prediction. Using a rigorous CRISP-DM pipeline on 119,390 records, we applied SMOTE exclusively to the training set and engineered six behavioral features to capture non-linear contradictions, such as the counter-intuitive 99.36% cancellation rate in non-refund deposits. We systematically benchmarked RF (bagging) against XGBoost (boosting) using stratified 5-fold cross-validation, hyperparameter optimization, and loss curve monitoring. Results demonstrate that XGBoost structurally outperforms RF in minority-class detection, achieving superior Recall (0.6659), F1-Score (0.6607), and AUC-ROC (0.8657), with significantly lower variance (0.0035). Conversely, RF exhibited higher Precision (0.6588) and better cross-validation stability during hyperparameter search. Crucially, feature importance analysis revealed a structural divergence: RF prioritized temporal variables (lead_time), while XGBoost emphasized behavioral commitment signals (parking, special requests). These findings confirm that gradient boosting’s sequential residual-correction mechanism is inherently more robust than variance-reduction bagging for imbalanced tabular data containing complex, non-linear anomalies. The proposed leakage-free framework not only resolves methodological flaws in prior studies but also provides a reliable, proactive risk-scoring foundation for integrating real-time decision support systems in production environments.

Synergizing Historical Similarity and Probabilistic Attributes: A Dual-Engine Machine Learning Model for Subsidized Housing Credit Risk

2025-12-29T13:53:39+07:00

Subsidized mortgage programs require objective and efficient credit approval processes to mitigate subjective biases and manual inefficiencies inherent in traditional evaluations. This study proposes a web-based decision support system leveraging a hybrid ensemble architecture that integrates the K-Nearest Neighbor (KNN) and Naïve Bayes Classifier (NBC) algorithms to assess the creditworthiness of subsidized mortgage applicants at Raja Batu Residence. The proposed framework utilizes KNN for historical data similarity mapping and NBC for probabilistic attribute evaluation, combining their predictions through a Soft Voting Aggregation mechanism to enhance stability. The system's performance was rigorously evaluated using optimized classification metrics and the System Usability Scale (SUS) involving 20 end-users. Empirical results demonstrate outstanding classification performance, with the optimized KNN model achieving an overall accuracy of 93.33% (yielding only 2 false positives and 3 false negatives) and the Naïve Bayes model achieving 92.00% accuracy (yielding 4 false positives and 2 false negatives). This high classification accuracy ensures a well-balanced confusion matrix with a minimized risk profile, aligning effectively with the prudent risk management required for government-subsidized allocations. Furthermore, the deployed web application achieved an "Excellent" usability score of 82.75, confirming its practical viability for non-technical administrative staff. Ultimately, the hybrid integration of KNN and NBC successfully streamlines the credit evaluation workflow, minimizing subjective bias while providing a reliable, highly accurate, and user-friendly tool for housing developers.

Beyond Intensity Gradients: A Lightweight Client-Side Morphological Framework for Robust Binary Image Segmentation

2025-12-29T13:54:52+07:00

Edge detection in binary images frequently suffers from contour fragmentation when utilizing conventional gradient operators (e.g., Sobel, Canny), as they inherently rely on continuous intensity variations absent in discrete binary data. To address this, this study proposes a lightweight, client-side morphological gradient framework designed to structurally and topologically extract object boundaries based purely on spatial geometry. A modular web-based computational pipeline was developed using a PHP-JavaScript architecture, integrating adaptive Otsu binarization, dynamic kernel configurations, and core morphological computations. For rigorous algorithmic and engineering validation, the system was benchmarked against the conventional Canny detector using a dataset of 50 botanical images, supplemented by comprehensive white-box and black-box software testing involving 20 participants across 20 distinct test cases. Experimental results demonstrate that the proposed morphological framework achieves superior topological fidelity and boundary continuity. It secured an optimal F1-score of 0.9006 using a 3×3 disk structuring element, significantly outperforming the Canny detector’s F1-score of 0.7761. Furthermore, the method exhibits highly parallelizable efficiency, executing edge extraction at a mean runtime of 18.2 ms, more than twice as fast as the multi-stage Canny operator (42.5 ms). Software evaluations confirmed exceptional system reliability, yielding a 97.4% functional success rate, 95% overall module coverage, and an excellent usability score of 4.58/5.0. Ultimately, this research proves that shifting the paradigm from intensity differentiation to mathematical morphology offers a highly precise, robust, and computationally lightweight solution for real-time, web-native binary image segmentation.

Majority-Class Collapse in Low-Resource E-Commerce Emotion Mining: Diagnostic Evaluation of LSTM on Imbalanced and Linguistically Noisy Indonesian Reviews

2025-12-29T13:55:28+07:00

Product reviews on e-commerce platforms such as Shopee provide valuable emotional information that can support customer behavior analysis and business decision-making. However, automatically identifying fine-grained emotions remains challenging because user-generated reviews often contain informal language, abbreviations, spelling variations, and severe class imbalance. Although Long Short-Term Memory (LSTM) networks have demonstrated strong capabilities in sequential text modeling, their robustness under highly imbalanced and low-resource real-world conditions remains insufficiently investigated. This study evaluates an LSTM-Word2Vec architecture for multi-class emotion classification (positive, neutral, and negative) using 306 Indomie product reviews collected from Shopee through web crawling. The proposed methodology includes comprehensive text preprocessing, custom Word2Vec semantic embedding generation, LSTM-based classification, and performance evaluation using Accuracy, Precision, Recall, and F1-Score. To further quantify prediction deviations, numerical error metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) are also employed. Experimental results reveal a pronounced majority-class collapse, where the model achieved only 31% accuracy by predicting nearly all testing samples as the negative class. Detailed diagnostic analysis indicates that this failure is primarily caused by the combined effects of severe class imbalance, the limited dataset size, and noisy labels generated through heuristic proxy annotation. These findings demonstrate that conventional LSTM architectures are highly vulnerable to noisy, colloquial marketplace language when adequate data quality is unavailable. Consequently, this research establishes a realistic baseline for e-commerce emotion mining and emphasizes that model complexity alone cannot overcome fundamental data limitations.

Beyond the Black Box in Computer Vision: A Traceable Architecture for Reproducible Canny Contour Identification

2025-12-29T14:01:06+07:00

While modern edge detection algorithms are pivotal in advanced computer vision pipelines, conventional implementations typically operate as static, transient black-boxes, severely hindering experimental replication and parameter traceability. This study addresses this methodological bottleneck by developing a structured, web-based contour identification system designed to enforce configuration management within the Canny framework. The decoupled architecture fuses a PHP-driven interface with a high-performance Python backend and a persistent relational database layer, allowing for the deterministic tracking of localized hysteresis thresholds, aperture scales, and spatial gradient vectors. Empirical evaluations were executed on a curated dataset comprising high-contrast object morphologies, characterized by distinct structural boundaries and varied illumination backgrounds to rigorously test edge degradation. The transition from linear Manhattan approximations to an isotropic Euclidean space ( gradient norm) yields single-pixel edge localization sharpness and unbroken contour continuity. Quantitatively, this mathematical refinement achieves a peak F-measure boundary accuracy of 0.91 and a Pratt’s Figure of Merit of 0.895, albeit introducing a 16.8% latency overhead. Furthermore, a multi-factor Analysis of Variance (ANOVA) robustly rejects the null hypothesis, confirming that parameter interactions significantly dictate contour fidelity (). The primary contribution of this research is the transformation of a heuristic vision task into a deterministic, database-backed ecosystem. By embedding an explicit audit trail for every processing trace, this framework provides computer vision practitioners with a rigorous instrument for quasi-quantitative comparative analysis, establishing a transparent benchmark for reproducible boundary extraction in downstream visual recognition tasks.

A Dual-Pipeline Imbalance-Robust Framework for SMS Spam Detection: Achieving Flawless Precision via SMOTE-Augmented Ensembles with Rigorous Statistical Validation

2026-06-18T13:04:04+07:00

The rapid proliferation of digital communication has exponentially increased the volume of Short Message Service (SMS) spam, exposing mobile users to systemic convenience disruptions, productivity drops, and severe financial losses through sophisticated fraudulent schemes. To construct a highly dependable filtering mechanism, this study presents a rigorous dual-pipeline machine learning framework that systematically addresses the challenges of class imbalance in statistical text mining. Utilizing a verified dataset of 5,572 Indonesian-context short messages, the raw textual corpus is subjected to uniform case normalization, structural URL extraction, and character filtering before feature projection via Term Frequency–Inverse Document Frequency (TF-IDF) vectorization. To overcome the inherent accuracy paradox of skewed class distributions, the experimental design evaluates a baseline pipeline (imbalanced data) against a synthetic data augmentation pipeline leveraging the Synthetic Minority Oversampling Technique (SMOTE) across four distinct classifiers: Logistic Regression, Naive Bayes, Linear Support Vector Machine (Linear SVM), and Random Forest. Empirical results demonstrate that while the baseline Linear SVM serves as the optimal standalone model for overall balance, achieving a peak accuracy of 98.11% and a dominant F1-Score of 92.83%, the SMOTE-augmented Random Forest configuration yields an exceptional high-security alternative by securing a flawless 100.00% precision envelope alongside an 83.89% recall rate. Advanced post-hoc evaluations including McNemar's statistical significance tests (, for Random Forest), qualitative error analyses of semantic edge cases, and runtime profiling confirm that the developed architecture establishes a highly scalable, mathematically verified, and low-latency solution suitable for integration into real-time telecom filtering gateways.