Publications.
Peer-reviewed work spanning object detection, speech AI and software engineering — from military aircraft at 94% mAP to maintainability prediction. Plus 28 review records across 4 international journals.
Most active in 2024.
Two papers in 2024 — the MFCC speech study and the YOLOv7/v8 conference precursor. One forthcoming in 2026.
Strong upswing — 15 citations in 2025–26.
Most citations land on the YOLO / object-detection track. 2026 is ongoing (*partial year).
Published work.
// 6 RECORDS · NEWEST FIRST-
—
IN PRESS
Software Maintainability Prediction Using Static Code Metrics and Machine Learning FORTHCOMING
Currently in publication process
Crossing my research background with day-to-day ABAP / Fiori work: can static code metrics alone predict how maintainable a module will be? Trains ML classifiers on metric vectors to forecast maintainability tiers — aimed at teams who need a triage signal before a full review. Full abstract and results publish with the paper.
-
2025
/ 01
Detection of Military Aircraft Using YOLO and Transformer-Based Object Detection Models in Complex Environments
Bilişim Teknolojileri Dergisi · Vol. 18, No. 1, pp. 85–97 · Gazi University
Benchmarks YOLOv7, YOLOv8 and RT-DETR on a 19,514-image dataset spanning 43 aircraft models, shot from varied angles against urban, rural and coastal backgrounds. After hyperparameter tuning and class-imbalance analysis, YOLOv8 leads at 94.0% mAP / 88.1% recall, RT-DETR follows with 92.7% mAP and the highest recall (90.4%), YOLOv7 trails at 90.2% / 82.7%. The paper gives a practical read on where each family breaks down in cluttered defense scenes.
mAP · 94.0% Recall · 88.1% Classes · 43 Images · 19,514 -
2024
/ 02
A Modified MFCC-based Deep Learning Method for Emotion Classification from Speech
International Advanced Researches and Engineering Journal · Vol. 8, No. 1, pp. 33–42
Speech carries emotion, not just vocabulary — a signal worth decoding for richer human–machine interaction. Using the TESS dataset from the University of Toronto, the paper extracts MFCC feature maps and feeds them into a purpose-built CNN, benchmarked against an LSTM baseline. The MFCC + CNN pipeline reaches 99.5% accuracy on emotion classification, outperforming prior work on the same set.
Accuracy · 99.5% Dataset · TESS Feature · MFCC Model · CNN vs. LSTM -
2024
/ 03
YOLOv7 ve YOLOv8 Algoritmalarının Değerlendirilmesi: Savaş Uçaklarının Tespiti İçin Performans Analizi
Conference paper · 43-class, 19,514-image fighter-jet dataset
Head-to-head evaluation of YOLOv7 and YOLOv8 on military aircraft detection across complex backgrounds and varied angles. Reported on precision, recall and mAP, YOLOv8 edges YOLOv7 at 0.940 vs. 0.902 mAP. Both detect jets reliably in cluttered scenes — the paper positions YOLOv8 as the stronger candidate for downstream defense-AI pipelines. Conference precursor to the 2025 journal study.
YOLOv8 mAP · 0.940 YOLOv7 mAP · 0.902 Classes · 43 -
2023
/ 04
Sentiment Analysis Based on Machine Learning Methods on Twitter Data Using oneAPI
International Conference on Contemporary Academic Research · Accelerated with Intel oneAPI during ambassador term
Tweets turned into signal: using the Sentiment140 corpus, the paper runs a clean-then-classify pipeline with Bernoulli Naive Bayes, Linear SVM, Logistic Regression, LSTM and CNN — the last two combined in a hybrid. Scored by F1 and accuracy, the deep models top out at 0.85 (CNN) and 0.83 (LSTM), beating the classical baselines. The whole stack runs on Intel oneAPI toolkits — the work grew out of my oneAPI Student Ambassador term.
CNN · 0.85 LSTM · 0.83 Logistic · 0.83 Stack · Intel oneAPI -
2022
/ 05
Using Image Processing and Machine Learning Algorithms for the Detection of Surface Cracks
SIVAS International Conference on Scientific and Innovation Research · Vol. 1, No. 1, pp. 1201–1214 · IKSAD Institute
Manual crack inspection on buildings and roads is slow, costly and subjective. This work replaces it with an automated pipeline: morphological image processing and Hough-line transform extract crack features, then Naive Bayes, Random Forest, KNN and MLP classify. The winning combo — Hough-line + Random Forest — hits 97% accuracy, landing as a credible low-cost alternative for infrastructure condition monitoring.
Accuracy · 97% Best feature · Hough-line Best model · Random Forest
Reviewing for.
// 28 REVIEWS · 21 MANUSCRIPTS · 4 JOURNALSEngineering Research Express
IOP Publishing
Active since 2024 · YOLO, defect detection, fault diagnosis
Measurement Science and Technology
IOP Publishing
Active since 2023 · YOLO variants, infrared imaging, AGV
Physica Scripta
IOP Publishing
Since 2024 · Smoke / fault detection in transmission
Indian Journal of Science & Technology
Indian Society for Education and Environment
Since 2026 · Aircraft / Vision Transformer detection
International Advanced Researches and Engineering Journal
Dergipark · Open access
Since 2024 · Game engines, applied engineering ML
Selected recent reviews.
// LAST 12 MONTHS · ANONYMIZED PER POLICYOpen Google Scholar.
Live citation graph, full author list, and any newer drops not yet on this page.
Open ResearchGate.
Full PDFs, co-author network and discussions on each paper.