Smart Traffic Control System Using YOLO Algorithm

Appears in:

ismla 26(1):

Authors:

Lokesh S , Vismaya R , Shravan Hiremath , Sanmith Adwik , V Brunda

Published:

1st Aug 2026

Abstract:

Traffic congestion is one of the most serious challenges faced by developing countries, significantly affecting economic productivity and environmental sustainability. Most conventional traffic signal systems operate using fixed-time control strategies and do not adapt to real-time road conditions, leading to traffic congestion and inefficient utilization of road capacity. To address this issue, this study proposes an adaptive traffic signal control system that dynamically adjusts signal timings based on real-time traffic density. The proposed system employs a YOLO-based vehicle detection model integrated with OpenCV to analyze live traffic video streams and perform lane-wise vehicle counting. Based on the detected traffic density, green signal durations are computed dynamically to optimize traffic flow across multiple lanes. Experimental evaluation demonstrates that the system achieves approximately 93% vehicle detection accuracy and reduces average waiting time by nearly 30% compared to traditional fixed-time signal control systems. The results indicate improved traffic distribution, reduced idle time, and enhanced overall junction efficiency, demonstrating the practical applicability of the proposed approach for intelligent transportation systems and smart city environments.

---

BERTopic-Based Topic Modeling and Thematic Discovery in Long-Form Narrative Text

Appears in:

ismla 26(1):

Authors:

Haragopal V.V. , Sarojamma B. , I.B.N HimaBindu

Published:

4th Jun 2026

Abstract:

With the increasing amount of digital text data available today, the demand for Natural Language Processing techniques is growing significantly. Topic modeling is a NLP technique for automatically identifying topics existing in a large corpus of text and deriving hidden patterns represented by that document collection, hence facilitating improved decision-making. The purpose of the present work is to explore the major topics of the renowned book “Autobiography of a Yogi”, written by Paramahansa Yogananda, an eloquent orator and a profound spiritual master. To accomplish the study, the most popular neural topic model ‘BERTopic’ was employed on the book. As a result, a number of intriguing topics are extracted, that are especially useful for those researchers and scholars delving into the complexities of the book as well as those interested in spirituality, Indian philosophy, the life journey and teachings of Paramahansa Yogananda.

---

Enhancing Credit Card Fraud Detection under Severe Class Imbalance using Cost-Sensitive Learning and Threshold Optimization

Appears in:

ismla 26(1):

Authors:

Vu Ngoc Thanh Sang

Published:

1st Jun 2026

Abstract:

INTRODUCTION: Credit card fraud detection remains challenging because fraudulent transactions are rare, fraud patterns evolve over time, and precision–recall trade-offs vary under different class prevalences. While synthetic oversampling methods such as SMOTE are widely used, they may alter the observed feature distribution and complicate deployment interpretation. OBJECTIVES: This study develops a cost-sensitive and decision-threshold-calibrated fraud detection framework that preserves the original data distribution under severe class imbalance. METHODS: The framework combines cost-sensitive XGBoost and a Multi-Layer Perceptron trained with weighted Binary Cross-Entropy and Focal Loss. Hyperparameters are tuned using Optuna within a leakage-conscious validation protocol, and class imbalance is handled through scaleaware weighting rather than synthetic resampling. Decision thresholds are selected using minority-class F1 and an illustrative amount-aware cost criterion. SHAP analysis and a chronological split of the 2013 dataset are used to examine transformed-feature auditability and near-future generalization. RESULTS: On the imbalanced 2013 dataset, the optimized XGBoost model improves test-set PR-AUC from 0.7809 to 0.8815 and minority-class F1 from 0.7919 to 0.8497, with false positives decreasing from 21 to 13 on a test partition containing 98 fraud cases. On the balanced 2023 dataset, ranking performance is near-saturated and threshold selection mainly reduces false alarms. Under the illustrative assumption CFP = 1, amount-aware thresholding yields a lower simulated cost than the F1-optimized threshold, but with substantially more false positives. Chronological validation on the 2013 dataset yields lower PR-AUC than random stratified evaluation. CONCLUSION: The results suggest that fraud detection under class imbalance benefits from separating ranking optimization from decision-threshold selection. However, cost-aware and temporal findings should be interpreted as benchmark-based, deployment-motivated analyses rather than direct evidence of operational deployment readiness.

---

Lightweight and Real-Time Object Detection on Edge Devices: A Unified Framework for Resource-Constrained Environments

Appears in:

ismla 26(1):

Authors:

Hewa Majeed Zangana

Published:

12nd May 2026

Abstract:

Advances in edge computing have heightened the demand for object detection models capable of running efficiently on devices with constrained computational resources. This paper presents a robust hybrid detection framework that integrates template matching with Faster R-CNN to enhance detection accuracy in challenging conditions, such as occlusion, low lighting, and motion blur, while maintaining reasonable computational efficiency. Unlike conventional cloud-based detection, our approach reduces latency and improves data privacy. On the LASIESTA dataset, the proposed method achieves a mean Average Precision (mAP) of 88.2% at IoU 0.5 and 74.6% at IoU 0.75, outperforming Faster R-CNN by 4.3% in precision and 3.6% in recall. Although inference time increases modestly by 6 ms/frame compared to Faster R-CNN alone, the hybrid method consistently delivers superior robustness. While our implementation focuses on performance evaluation on a workstation, the framework design can be adapted for deployment on heterogeneous devices through additional optimization steps such as pruning and quantization. These findings demonstrate that combining classical localization techniques with deep learning models yields a practical and effective solution for real-time detection in resource-constrained environments.  

---

A Unified Hand-Landmark-Based Deep Learning Framework for Static and Dynamic Vietnamese Sign Language Recognition

Appears in:

ismla 26(1):

Authors:

Linh Duong Thanh, Don Pham Kim

Published:

13rd Apr 2026

Abstract:

Sign language recognition plays a crucial role in supporting communication between deaf communities and hearing individuals. In particular, Vietnamese Sign Language (VSL) recognition remains a challenging task due to the complexity of hand gestures and limited available datasets. However, most existing approaches address static gestures and dynamic sign phrases as separate recognition problems, often employing different feature representations and independent processing pipelines. This fragmentation increases system complexity and limits scalability for real-time sign language applications. This study proposes a unified hand-landmark-based deep learning framework for recognizing both static and dynamic VSL gestures within a single integrated system. The system begins by detecting hand landmarks through the MediaPipe hand tracking pipeline. Based on temporal motion analysis of landmark sequences, a gesture routing mechanism automatically determines whether the input corresponds to a static gesture or a dynamic sign phrase. Static gestures are classified using a convolutional neural network (CNN), while dynamic gestures are processed using a long short-term memory (LSTM) network to capture temporal dependencies. Experiments were conducted on a VSL dataset consisting of 23 static hand signs and 4 dynamic sign phrases collected from multiple participants with variations in hand shape and gesture execution style. The system performance is evaluated using accuracy, precision, recall, and F1-score. Experimental results demonstrate that the proposed framework achieves an average recognition accuracy of 92% for static gestures and 88.4% for dynamic sign phrases, outperforming traditional machine learning baselines. The proposed system provides a practical and efficient solution for VSL recognition and has potential applications in real-time assistive communication systems.

---

The Relationship Between the Exchange Rate and the General Index of the Istanbul Stock Exchange (BIST100) by Machine Learning Application

Appears in:

ismla 26(1):

Authors:

Adil R N , Abdul Razzaq H H

Published:

10th Apr 2026

Abstract:

The purpose of this research is to understand the type of relationship between the exchange rate for the period (2020-2022) and the Istanbul Stock Exchange index by using machine learning application. This research stems from the assumption that exchange rate fluctuations directly affect the overall market indicator. The interval model of self-regression distribution is used in general market indicators, and the GRANGER test was used to demonstrate the cause-effect relationship among economic variables. Turning to the market index, the study concluded that the exchange rate explains the changes in the general market index (best 100) (97%) and (3%) for variables other than the standard model, and there is a long-term joint integration relationship between the research variables, the result is that the exchange rate explains the changes in the general market index (best 100) (97%) and (3%) for variables other than the standard model.

---

Appears in:

ismla 26(1):

Authors:

None

Published:

25th Feb 2026

Abstract:

---

Appears in:

ismla 26(1):

Authors:

None

Published:

14th Jan 2026

Abstract:

---

Appears in:

ismla 26(1):

Authors:

None

Published:

5th Jan 2026

Abstract:

---

Quality Aware Content-Based Image Retrieval Using QIM and Deep Learning for Big Visual Data

Appears in:

ismla 25(1):

Authors:

Amit Phadikar

Published:

2nd Dec 2025

Abstract:

Modern technology has made storing, sharing, and organizing huge amounts of data simple through the Internet of Things. Search engines and query-based retrieval databases made access to relevant data easy through ranking and indexing based on stored content. In this paper, a secure CBIR scheme based on watermarking is proposed. Firstly, the image owner embeds the watermark in the image using quantization index modulation (QIM) in the luminance (Y) color space. The watermarked images are then uploaded to the cloud server, which extracts image feature vectors. In this article, features derived from a pre-trained network model from a deep-learning convolutional neural network trained for large image classification have been used for the retrieval of similar images. The image similarity is calculated using Euclidean distance, and the precision (P) is used as the performance measure of the model that achieved nearly 100%. Extensive experiments are carried out, and assessment results reveal the outperforming result of the proposed technique compared to other related schemes. The scheme can be used in many applications that need CBIR, such as digital libraries, historical research, fingerprint identification, and crime prevention.

---