Prof. Dr. M. Ali Akcayol
Gazi Üniversitesi
Bilgisayar Mühendisliği Bölümü
akcayol@gazi.edu.tr,  maakcayol@gmail.com


Fotoğraflar
Ana Sayfa|Yayınlar|Projeler|Araştırma Alanları|Dersler|Öğrenciler|Akademik Deneyim|Duyurular|İletişim


KISA ÖZGEÇMİŞ (Özgeçmiş)
Lisans, Yüksek Lisans ve Doktora derecelerini Gazi Üniversitesinden sırasıyla 1993, 1998 ve 2002 yıllarında almıştır. Doktora tezinde, bulanık mantık ve yapay sinir ağlarının dinamik sistemlere uygulanması konusunda çalışmıştır.

2004 yılında doktora sonrası araştırma için Michigan State Üniversitesinde bulunmuştur.

Ulusal ve uluslararası kuruluşlardan destek alan projelerin yapıldığı Büyük Veri ve Mobil Sistemler Laboratuvarının yöneticisidir. Yürütücüsü olduğu projeler, Gazi Üniversitesi, TÜBİTAK, Bilim Sanayi ve Ticaret Bakanlığı, TAİ ve Huawei başta olmak üzere çeşitli kamu ve özel sektör kuruluşları tarafından desteklenmiştir.

Çok sayıda uluslararası ve ulusal dergide editörlük/alan editörlüğü ve hakemlik yapmaktadır. Dünyada saygın konferanslarda oturum başkanlığı, düzenleme kurulu üyeliği ve bilim kurulu üyeliği yapmıştır. ACM profesyonel üyesidir.

Çalışma konuları arasında, yapay zeka, derin öğrenme, büyük veri analitiği, öneri sistemleri, zeki optimizasyon teknikleri, mobil kablosuz ağlar ve akıllı binalar yer almaktadır.
           


GÜNCEL PROJELER (Tüm projeler)
Çok amaçlı hibrit otonom dron tasarımı ve prototip imalatı
TÜBİTAK 3211162. (Danışman)
İniş sahası tespit sistemi tasarımı ve prototip imalatı
TÜBİTAK 3211383. (Danışman)
Planlama süreçlerine kentsel ısı adası etkisi azaltımının entegrasyonu için bir model: Yerel iklim bölgesi temelli morfolojik yaklaşım
TÜBİTAK 1001. (Araştırmacı)
İnternette heterojen veri kaynaklarından veri toplanması, doğrulanması ve sorgulanması
TÜBİTAK 2244 Sanayi Doktora Programı (Huawei Telekomünikasyon Ltd. Şti.), 118C127. (Proje Yürütücüsü)


SON YAYINLAR (Tüm yayınlar) 
Machine learning-based comparative study for heart disease prediction
Güllü M., Akcayol M.A., Barışçı N.
Advances in Artificial Intelligence Research, Vol.2(2), pp.51-58, 2022. (Selected from ICAIAME 2022)
Abstract | pdf
Heart disease is one of the most common causes of death globally. In this study, machine learning algorithms and models widely used in the literature to predict heart disease have been extensively compared, and a hybrid feature selection based on genetic algorithm and tabu search methods have been developed. The proposed system consists of three components: (1) preprocess of datasets, (2) feature selection with genetic and tabu search algorithm, and (3) classification module. The models have been tested using different datasets, and detailed comparisons and analysis were presented. The experimental results show that the Random Forest algorithm is more successful than Adaboost, Bagging, Logitboost, and Support Vector machine using Cleveland and Statlog datasets.
Trust-chain-based certificate revocation control in autonomous vehicle networks
Erceylan G., Akcayol M.A.
IEEE 5th International Conference on Information and Communications Technology, Universitas AMIKOM Yogyakarta, Indonesia, August 24-25, 2022.
Abstract
One of the biggest problems in V2X communication is the inclusion of faulty or illegal nodes in the network. In V2X systems using PKI, the primary approach is to revoke the certificates of malicious and compromised vehicles and add them to a CRL. The issue of CRL distribution from the central server without high latency and high traffic is a subject of research. In this paper, a semi-distributed certificate revocation control system called trust-chain is proposed, which reduces the demand for CRL servers. In the simulation results, the number of requests to the central CRL server is reduced by 35% on average using the proposed semi-distributed structure. Additionally, in cases where vehicles cannot communicate with the RSU or central server, vehicles can communicate securely owing to the semi-distributed framework of the proposed system.
Deep learning based video event classification
Gençaslan S., Utku A., Akcayol M.A.
Journal of Polytechnic, (Accepted).
Abstract
In recent years, due to the growth of digital libraries and video databases, automatic detection of activities from videos and obtaining patterns from large datasets have come to the fore. Object detection from image is used as a tool for various applications and is the basis of video classification. Objects in videos are more difficult to identify than in a single image, as the information in videos has a time continuity constraint. Following the developments in the field of computer vision, the use of open source software packages for machine learning and deep learning and the developments in hardware technologies have enabled the development of new approaches. In this study, a deep learning-based classification model has been developed for the classification of sports branches on video. In the model developed using CNN, transfer learning has been applied with VGG-19. Experimental studies on 32827 frames using CNN and VGG-19 models showed that VGG-19 has a more successful classification performance than CNN with an accuracy rate of 83%.
SUST-DDD: A real-drive dataset for driver drowsiness detection
Yılmaz K.E., Akcayol M.A.
IEEE 31st Conference of the Open Innovations Association FRUCT, Helsinki, Finland, April 27-29, 2022.
Abstract | pdf
Driver drowsiness is one of the most important factors in traffic accidents. For this reason, systems should be developed to detect drowsiness early and to warn the driver by examining the driver or driving situations. These developed systems play an important role in preventing accidents. Three techniques are used to detect drowsiness: ’Based on Vehicle Parameters’, ’Based on Physiological Parameters’ and ’Based on Behavioral Parameters’. In this study, the studies on fatigue detection systems were examined and a literature study was presented and the techniques used were examined. The deep learning methods used in the studies were also examined and presented. Finally, the data sets used in the studies were compared and the general results were shared.
Deep learning-based forecasting of cancellation, delay and orientation on flights
Ayaydın A., Akcayol M.A.
Journal of Informatics Technologies, Vol.15(3), pp.239-249, 2022. (Selected from ICI-CS2021)
Abstract | pdf
In this study, three different methods from machine learning and deep learning have been implemented for preventing financial and moral losses that may occur as a result of delays in flights and to take necessary precautions by predicting the flight delay in advance, which are a serious problem in the aviation industry. Deep recurrent neural network (DRNN), long-short term memory (LSTM), and random forest (RF) have been extensively tested and compared employing a real data set covering 368 airports across the world with relevancy the success rate of forecasting of delay on flights. The experimental results showed that the LSTM model had a higher success rate of 96.50% at the recall level than the others.
Machine learning-based comparative study for heart disease prediction
Güllü M., Akcayol M.A., Barışçı N.
International Conference on Artificial Intelligence and Applied Mathematics in Engineering, Baku, Azerbaijan, May 20-22, 2022.
Development of connected vehicle technology in Turkey
Erceylan G., Akcayol M.A.
International Symposium on Innovative Approaches in Smart Technologies, Ankara, Turkey, May 28-29, 2022.
A comprehensive review of image denoising with deep learning
Yapıcı A., Akcayol M.A.
International Journal of Advances in Engineering and Pure Sciences, Vol.34(1), pp.65-69, 2022.
Abstract | pdf
In daily life and scientific searches, the need for real-like and denoised images is increasing. But images are distorted by noise, resulting in lower visual image quality. For this reason, noise removal studies are carried out on images to increase the quality. Until now, various methods have been proposed to decrease noise, each technique has different advantages. This paper gives information about the methods that achieve the best results in their field and summarizes the studies about traditional denoising and deep learning based denoising methods in the field of noise reduction in video and images and compares the studies with each other. Researches show that experiments focus on the case of additive white Gaussian noise. Traditional noise removal methods, machine learning methods, deep learning methods and other mathematical methods have been used in image denoising problem over time, and deep learning methods achieve more successful results. However, according to the obtained data, it has been seen that the studies on training the model without having the original image pairs were insufficient and a single method could not be successful at different noise levels. In future studies, it is necessary to focus on how to remove the noise in real-life images.
A novel Web ranking algorithm based on pages multi-attribute
Baker M.R., Akcayol M.A.
International Journal of Information Technology, DOI: 10.1007/s41870-021-00833-5, 2022.
Abstract | pdf
The size of the Internet is rapidly increasing. It has become a necessity to access information on the web correctly for a short time. For this reason, search engines have arisen out of meeting this need. In this study, we propose a ranking algorithm based on page multi-attribute(PMARank). The proposed algorithm uses a novel index calculation system that acts as a pre-rank process for web pages. In the ranking procedure, the featured meta-tag of a page and its contents were extracted to locate words as ranking features. The proposed web ranking algorithm has been compared with PageRank (PR) and Hyperlink-Induced Topic Search (HITS) algorithms. Experimental results show that the proposed ranking algorithm performs better than PR and HITS algorithms according to user clickstreams on the search results page.
Bird species classification using deep learning: a comparative study
Bilgin M.M., Özdem K., Akcayol M.A.
Journal of Polytechnic, (Accepted).
Abstract | pdf
Studies to classify bird species on the basis of images are very difficult due to both the abundance of colors and patterns in the image, and their very close visual characteristics. In this study, six different deep learning models have been applied for the classification of bird species and the experimental results have been compared comprehensively. A dataset named 250 Bird Species, which includes a total of 31316 bird images with 225 bird species, was used as dataset. In the study, 1125 images have been used for the test and 1125 images for the validation. The comparison of VGG16, ResNet50, ResNet152V2, InceptionV3, MobileNet and DenseNet121 deep learning models have been made on the dataset respectively, according to the accuracy, precision, recall and F1-score values. In experimental studies, 94.6% accuracy value ??has been obtained with VGG16, 47.2% with ResNet50, 96.2% with ResNet152V2, 97.5% with InceptionV3, 96.9% with MobileNet and 98.2% with DenseNet121. DenseNet121 obtained the highest precision value as 0.99, sensitivity value as 0.99 and F1-score value as 0.99.
A new topic modeling based approach for aspect extraction in aspect based sentiment analysis: SS-LDA
Özyurt B., Akcayol M.A.
Expert Systems with Applications, Vol.168, 114231, April 2021.
Abstract | pdf
With the widespread use of social networks, blogs, forums and e-commerce web sites, the volume of user generated textual data is growing exponentially. User opinions in product reviews or in other textual data are crucial for manufacturers, retailers and providers of the products and services. Therefore, sentiment analysis and opinion mining have become important research areas. In user reviews mining, topic modeling based approaches and Latent Dirichlet Allocation (LDA) are significant methods that are used in extracting product aspects in aspect based sentiment analysis. However, LDA cannot be directly applied on user reviews and on other short texts because of data sparsity problem and lack of co-occurrence patterns. Several studies have been published for the adaptation of LDA for short texts. In this study, a novel method for aspect based sentiment analysis, Sentence Segment LDA (SS-LDA) is proposed. SS-LDA is a novel adaptation of LDA algorithm for product aspect extraction. The experimental results reveal that SS-LDA is quite competitive in extracting products aspects.
Prediction of the next time of an event with deep learning based model
Utku A., Akcayol M.A.
Journal of Polytechnic, Vol.24(1), pp.1-15, 2021.
Abstract | pdf
Studies have been going on for many years to predict the time before some events happen. Thus, it is aimed to minimize the damage that occurs when the event occurs or to maximize the benefit to be obtained. Studies on the prediction of subsequent events in many different areas, such as the prediction of the subsequent behavior of a customer, the prediction of the subsequent occurrence of natural disasters, the estimate of the number of future demands in a given time interval, are gradually increasing. However, in the literature, there is no successful study for predicting the time and type of event before the occurrence of crimes and emergency calls. Crime analysis is a field of research aimed at securing the threatened areas, reducing the rate of crime and saving law enforcement. High success is achieved with the use of up-to-date technologies in the efforts to resolve the crime shortly after it is committed. Similarly, emergency call analysis reduces response time and optimizes resource usage. In this study, a deep learning based prediction model for crime and emergency call analysis has been developed. With the developed model, the time of the next crime and the time of the next emergency call are predicted. The results obtained with the developed model has been compared with ARIMA which is one of the statistical time series prediction methods. Experimental results have shown that the developed deep learning-based model is more successful than ARIMA in forward-looking event time prediction.
Locality sensitive hashing based clustering for large scale documents
Özdem K., Akcayol M.A.
ACM International Conference on Mathematics and Artificial Intelligence, Chengdu, China, March 19-21, 2021.
Abstract | pdf
Nowadays, the size of data continues to increase more rapidly every day. Considering this situation, large-scale processing has become a very important issue in document clustering, due to its capability to organize large numbers of documents into few meaningful and consistent clusters. In this study, a dataset consisting of 390 English textbooks of a total size of 7.61 GB, is used for the clustering task. Locality sensitive hashing and k-shingles methods are used to obtain clusters with high quality. Clusters are evaluated using cluster validity indices. According to the experimental results, high-quality clusters have been obtained, with 0.88 and 0.79 for Silhouette and Davies–Bouldin scores, respectively.
A review of image denoising with deep learning
Yapıcı A., Akcayol M.A.
IEEE International Informatics and Software Engineering Conference, Ankara, Turkey, December 16-17, 2021.
Abstract | pdf
Satellite images can be corrupted by noise during image capture, transfer or due to bad environmental conditions. In daily life and scientific searches, the need for more accurate images are increasing. However, images are distorted by noise, resulting in lower visual image quality. For this reason, noise removal studies are carried out on images to increase the quality. Until now, various methods have been proposed to decrease noise and each technique have different advantages. This paper, summarizes the studies in the field of noise reduction in video and images and compares the studies with each other.
A new deep learning-based prediction model for purchase time prediction
Utku A., Akcayol M.A.
IEEE International Conference on Computer Science and Engineering, Ankara, Turkey, September 15-19, 2021.
Abstract | pdf
Nowadays, user behaviour analysis is gaining importance due to the increasing interest of users in e-commerce platforms. By providing users with a personalized shopping experience, customer satisfaction and sales rates can be increased. In this study, a deep learning based hybrid model has been developed for predicting the next purchase time. The developed model has been compared extensively with ARIMA and LSTM. Experimental results showed that the developed deep learning model has more successful than ARIMA and LSTM in predicting the next purchase time.
Bloom filter based graph database CRUD optimization for stream data
Hüseynli A., Akcayol M.A.
IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems, Cracow, Poland, September 22-25, 2021.
Abstract | pdf
This study has been prepared to set light on the performance difficulties encountered in large datasets on graph databases and to increase performance in Create, Read, Update, Delete (CRUD) operations with Approximate Membership Functions (AMF). For this purpose, the Bloom filter from the AMF family is proposed in a scalable structure. Neo4j commercial graph database was preferred in experimental studies as a graph data model. In the experimental studies, it has been observed that the proposed method for all CRUD operations produces better results than the BTREE indexing method used by the database. The proposed AMF method can be preferred for performance optimization in such databases.
Long short-term memory based query auto-completion
Qureshi A.R.A., Akcayol M.A.
IEEE International Conference on Electrical and Electronics Engineering, Antalya, Turkey, April 9-11, 2021.
Abstract | pdf
In this study, Long Short-Term Memory (LSTM) based Query Auto-Completion (QAC) has been proposed to generate a query completion list using input prefix. The performance of the QAC system has been evaluated by using the relevancy score, and the quality of the QAC generation system has been evaluated by using partial and complete matching strategies, success rate, normalized discounted cumulative gain, and mean average precision. The proposed LSTM based QAC system has been extensively tested using AOL and ORCAS datasets. According to experimental results, the performance of the proposed QAC system is more successful with the partial matching strategy. Also, the quality of the QAC generation list by the proposed QAC system is better on the complete matching strategy.
Automated machine learning platform
Selvi G., Dağ G., Dirican E.G., Aktay T., Aksu S.M., Özdem K., Akcayol M.A.
IEEE International Conference on Computer Science and Engineering, Ankara, Turkey, September 15-19, 2021.
Abstract | pdf
With the rapid development of information technologies and the widespread use of the internet, the volume and diversity of data has also increased. Meaningful information and important results can be obtained by processing this data, which is expressed with the concept of big data. In this study, a machine learning platform that can automatically learn from data sets with different data types and dimensions has been developed. When the dataset of any field is given as an input to the developed automatic machine learning platform, the most appropriate machine learning model is determined. With this platform, which has a Web interface that can be easily used by people who are experts in their field but do not have sufficient knowledge in the field of machine learning and data science, the most suitable machine learning model for the data set is suggested to the users and training and test results for different models can be obtained and compared. The experimental studies have shown that the developed platform is successful in knitting a machine learning model suitable for the dataset.
Thermal infrared colorization using deep learning
Çiftçi O., Akcayol M.A.
IEEE International Conference on Electrical and Electronics Engineering, Antalya, Turkey, April 9-11, 2021.
Abstract | pdf
Day by day the usage of infrared cameras has been increasing in the world. With the increasing use of thermal infrared cameras and images, especially in military, security and medicine, the need for coloring thermal infrared images to visible spectrum has arisen. In this study, a deep based model has been developed to generate visible spectrum images (RGB - Red Green Blue) from thermal infrared (TIR) images. In the proposed model, an encoder-decoder architecture with skip connections has been used to generate RGB images. KAIST-MS (Korea Advanced Institute of Science and Technology-Multispectral) dataset used for training and test the developed model. The experimental results extensively tested using Least Absolute Deviations (L1), Root Mean Squared Error (RMSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index Measure (SSIM).


SEÇİLMİŞ YAYINLAR (Tüm yayınlar) 
A new topic modeling based approach for aspect extraction in aspect based sentiment analysis: SS-LDA
Özyurt B., Akcayol M.A.
Expert Systems with Applications, Vol.168, 114231, April 2021.
Abstract | pdf
With the widespread use of social networks, blogs, forums and e-commerce web sites, the volume of user generated textual data is growing exponentially. User opinions in product reviews or in other textual data are crucial for manufacturers, retailers and providers of the products and services. Therefore, sentiment analysis and opinion mining have become important research areas. In user reviews mining, topic modeling based approaches and Latent Dirichlet Allocation (LDA) are significant methods that are used in extracting product aspects in aspect based sentiment analysis. However, LDA cannot be directly applied on user reviews and on other short texts because of data sparsity problem and lack of co-occurrence patterns. Several studies have been published for the adaptation of LDA for short texts. In this study, a novel method for aspect based sentiment analysis, Sentence Segment LDA (SS-LDA) is proposed. SS-LDA is a novel adaptation of LDA algorithm for product aspect extraction. The experimental results reveal that SS-LDA is quite competitive in extracting products aspects.
A weighted multi-attribute-based recommender system using extended user behavior analysis
Akcayol M.A., Utku A., Aydoğan E., Mutlu B.
Electronic Commerce Research and Applications, Vol.28, pp.86-93, 2018.
BibTeX | Abstract | pdf
A new weighted multi-attribute based recommender system (WMARS) has been developed using extended user behavior analysis. WMARS obtained data from number of clicked items in the recommendation list, sequence of the clicked items in recommendation the list, duration of tracking, number of tracking same item, likes/dislikes, association rules of clicked items, remarks for items. WMARS has been applied to a movie web site. The experimental results have been obtained from a total of 567 heterogeneous users, including employers in different sectors, different demographic groups, and undergraduate and graduate students. Using different weighted sets of the attributes’ parameters, WMARS has been tested and compared extensively with collaborative filtering. The experimental results show that WMARS is more successful than collaborative filtering for the data set that was used.
Calculation of electron energy distribution functions from electron swarm parameters using artificial neural network in SF6 and Argon
Tezcan S.S., Akcayol M.A., Ozerdem O.C., Dincer, M.S.
IEEE Transactions on Plasma Science, Vol.38(9), pp.2332-2339, 2010.
Abstract | pdf
This paper proposes an artificial neural network (ANN) to obtain the electron energy distribution functions (EEDFs) in SF6 and argon from the following: 1) mean energies; 2) the drift velocities; and 3) other related swarm data. In order to obtain the required swarm data, the electron swarm behavior in SF6 and argon is analyzed over the range of the density-reduced electric field strength E/N from 50 to 800 Td from a Boltzmann equation analysis based on the finite difference method under a steady-state Townsend condition. A comparison between the EEDFs calculated by the Boltzmann equation and by ANN for various values of E/N suggests that the proposed ANN yields good agreement of EEDFs with those of the Boltzmann equation solution results.
An educational tool for fuzzy logic controlled BDCM
Akcayol M.A., Çetin A., Elmas Ç.
IEEE Transactions on Education, Vol.45(1), pp.33-42, 2002.
BibTeX | Abstract | pdf
Fuzzy logic controllers (FLC) have gained popularity in the past few decades with successful implementation in many areas, including electrical machines’ drive control. Many colleges are now offering fuzzy logic courses due to successful applications of FLCs in nonlinear systems. However, teaching students a fuzzy logic controlled drive system in a laboratory, or training technical staff, is time consuming and may be an expensive task. This paper presents an educational tool for fuzzy logic controlled brushless direct current motor (BDCM), which is a part of a virtual electrical machinery laboratory project. The tool has flexible structure and graphical interface. Motor and controller parameters of the drive system can be changed easily under different operating conditions.
Application of adaptive neuro-fuzzy controller for SRM
Akcayol M.A.
Advances in Engineering Software, Vol.35(3-4), pp.129-137, 2004.
Abstract | pdf
In this paper, an adaptive neuro-fuzzy inference system (ANFIS) has been presented to speed control of a switched reluctance motor (SRM). SRMs have become an attractive alternative in variable speed drives due to their advantages such as structural simplicity, high reliability, high efficiency and low cost. But, the SRM performance often degrades for the machine parameter variations. The SRM converter is difficult to control due to its nonlinearities and parameter variations. In this study, to tackle these problems, an adaptive neurofuzzy controller is proposed. Heuristic rules are derived with the membership functions then the parameters of membership functions are tuned by ANFIS. The algorithm has been implemented on a digital signal processor (TMS320F240) allowing great flexibility for various real time applications. Experimental results demonstrate the effectiveness of the proposed ANFIS controller under different operating conditions of the SRM.
NEFCLASS-based neuro fuzzy controller for SRM drive
Akcayol M.A., Elmas Ç.
Engineering Applications of Artificial Intelligence, Vol.18(5), pp.595-602, 2005.
BibTeX | Abstract | pdf
Switched reluctance motor (SRM) is increasingly employed in industrial applications where variable speed is required because of their simple construction, ease of maintenance, low cost and high efficiency. However, the SRM performance often degrades for the machine parameter variations. The SRM converter is difficult to control due to its nonlinearities and parameter uncertainties. In this paper, to overcome this problem, a neuro fuzzy controller (NFC) is proposed. Heuristic rules are derived with the membership functions of the fuzzy variables tuned by a neural network (NN). The algorithm is implemented on a digital signal processor (TMS320F240) allowing great flexibility for various real time applications. Experimental results demonstrate the effectiveness of the NFC with various working conditions of the SRM.