Mahmut KARAASLAN

Mr. Mahmut KARAASLAN

orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 December 2024 in Sensors
Verified
Journal Article
Open access
PDF file
MKMahmut KARAASLAN
Mahmut KARAASLAN
BTBahaeddin Turkoglu
Bahaeddin Turkoglu
EKErsin KAYA
Ersin KAYA
TATunc Asuroglu
Tunc Asuroglu
Extracting behavioral information from animal sounds has long been a focus of research in bioacoustics, as sound-derived data are crucial for understanding animal behavior and environmental interactions. Traditional methods, which involve manual review of extensive recordings, pose significant challenges. This study proposes an automated system for detecting and classifying animal vocalizations, enhancing efficiency in behavior analysis. The system uses a preprocessing step to segment relevant sound regions from audio recordings, followed by feature extraction using Short-Time Fourier Transform (STFT), Mel-frequency cepstral coefficients (MFCCs), and linear-frequency cepstral coefficients (LFCCs). These features are input into convolutional neural network (CNN) classifiers to evaluate performance. Experimental results demonstrate the effectiveness of different CNN models and feature extraction methods, with AlexNet, DenseNet, EfficientNet, ResNet50, and ResNet152 being evaluated. The system achieves high accuracy in classifying vocal behaviors, such as barking and howling in dogs, providing a robust tool for behavioral analysis. The study highlights the importance of automated systems in bioacoustics research and suggests future improvements using deep learning-based methods for enhanced classification performance.

16 October 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
MKMahmut KARAASLAN
Mahmut KARAASLAN
AEArife Erdoğan
Arife Erdoğan
MEMümin Alper Erdoğan
Mümin Alper Erdoğan
SYSuzan Çelebioğlu Yüzügüldü
Suzan Çelebioğlu Yüzügüldü
HGHayriye Gönüllü
Hayriye Gönüllü
EKErsin Kaya
Ersin Kaya
Sepsis is a life-threatening syndrome that imposes a substantial burden on global healthcare systems. While artificial intelligence models have shown promise for early sepsis detection, the majority rely on continuous time-series data from intensive care units, limiting their applicability to emergency department (ED) settings where data are typically single-time, incomplete, and irregularly collected. This study presents a comprehensive framework for sepsis prediction using snapshot clinical data obtained at the point of initial patient admission. A retrospective dataset of 237 patients, labeled according to Sepsis-3 criteria, was used. Three feature subsets of increasing dimensionality were constructed based on discriminative power (Set 1: Minimalist with 5 features, Set 2: Balanced with 10, Set 3: Comprehensive with 21), and four imputation strategies were applied: Mean, Median, K-Nearest Neighbors (KNN), and a proposed Missing Gaussian Process (MGP) method that preserves the covariance structure between clinical variables. These combinations were evaluated across six classifiers comprising three classical machine learning models (Random Forest, SVM, XGBoost) and three deep learning architectures (CNN, LSTM, Attention). The MGP method achieved the lowest imputation error (RMSE: 32,270.65; R2: 0.83) and, when paired with tree-based classifiers, yielded an AUROC of 1.00 across all feature sets. In contrast, deep learning models exhibited marked performance degradation as dimensionality increased, particularly under univariate imputation. These results demonstrate that MGP-enhanced ensemble models constitute the most robust approach for ED-based sepsis prediction and that probabilistic imputation is critical for maintaining data integrity in incomplete clinical datasets.

21 May 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
MKMahmut KARAASLAN
Mahmut KARAASLAN
AAAhmed Abubaker
Ahmed Abubaker
EKErsin Kaya
Ersin Kaya
MSMohamed Saadeldin
Mohamed Saadeldin
Safe and autonomous movement capabilities in unmanned aerial vehicles (UAVs) aredirectly dependent on accurate and reliable perception of the environment. Depthinformation presents a fundamental requirement for critical tasks such as obstacleavoidance, path planning, and autonomous navigation. Although lidar and stereocamera systems provide high accuracy, they are not always applicable in UAV applications due to limitations such as cost, weight, and energy consumption. Therefore,monocular methods that estimate depth from a single RGB image offer a strong alternative for UAVs due to their low hardware requirements. However, the vast majorityof existing monocular depth estimation algorithms in the literature have been developed for ground-level driving scenarios or indoor environments and have not beenadequately evaluated under UAV-specific perspectives and environmental conditions.This study comprehensively compares three prominent monocular depth estimationmodels (DepthAnything V2, DistillAnyDepth, and Marigold) in the current literature on both real and simulation-based datasets. The models were evaluated in termsof error and accuracy metrics, as well as computational costs. Experimental resultsshow that the DepthAnything and DistillAnyDepth models generally offer higheraccuracy and stable performance, with DistillAnyDepth being the most suitable solution for real-time UAV applications due to its low latency. This study provides animportant comparative reference in the field of monocular depth estimation in UAVs