Browsing by Author "Bati, Cafer Tayyar"

Now showing 1 - 5 of 5

Evaluation of Machine Learning Hyperparameters Performance for Mice Protein Expression Data in Different Situations
(2021) Ser, Gazel; Bati, Cafer Tayyar
In this study, the aim was to assess the effect and significance of hyperparameters in four different datasets containing different values for observation numbers and variable counts with the machine-learning methods of support vector machines and artificial neural networks. With this aim, a dataset comprising 15 repeats of 77 protein levels from 38 healthy and 34 down syndrome mice was used. A total of 138 different models and model classification performance criteria were obtained from the datasets in the study comprising combinations of hyperparameters in machine-learning methods. Comparison of the models used criteria like accurate classification percentage, kappa statistic, mean absolute error and square root of mean error squares. According to performance criteria, the first dataset with 1080 observations x 77 variables had 71.30% accurate classification percentage for assumed parameters with the support vector machines polynomial kernel function, while changing the hyperparameter variables increased this rate to 99.44%. Similarly, the second dataset had 50.65% accurate classification percentage with the artificial neural network single hidden layer 2 neuron model, while changing the hyperparameter values increased this rate to 90.46%. In conclusion, in situations with low variable and observation numbers, the machine learning methods were determined to display lower performance. However, in datasets, it is very important for classification performance in artificial neural networks and support vector machines, especially polynomial and radial basis function kernel functions, to set hyperparameters according to the dataset. In situations with low variable numbers, especially, the effect of hyperparameters was determined to gain importance.
Evaluation of Parameter Estimation Methods for Determination of Covariance Structure in Repeated Data With Equal and Unequal Time Intervals
(Parlar Scientific Publications (p S P), 2018) Bati, Cafer Tayyar; Ser, Gazel
This study was completed with the aim of researching the effect of repeated measurement data with equal and unequal time intervals on the selection of the covariance structure and the parameter estimation methods of maximum likelihood (ML) and restricted maximum likelihood (REML). With this aim, blood glucose values took from the tail vein of 63 healthy rats administered plant extracts at varying rates over 21 days were used. Accordingly, while the most appropriate covariance structure was determined as Factor Analitic (FA(1)) for the data set at equal time intervals, the most appropriate covariance structure was determined as ANTE (1) for the dataset with unequal time intervals. There was no apparent difference difference determined for the performance of the parameter estimation methods of ML and REML for both data sets. In conclusion, while it is possible to try all homogeneous and heterogeneous variance-covariance structures to select the appropriate variance-covariance structure for equal time interval data sets, priority should be given to structures that take account of the time interval for unequal time interval data sets.
Improved Sheep Identification and Tracking Algorithm Based on Yolov5+sort Methods
(Springer London Ltd, 2024) Bati, Cafer Tayyar; Ser, Gazel
This research emphasises the importance of sheep identification and tracking in precision livestock farming and investigates the use of deep learning techniques for this purpose. Since traditional identification methods are time consuming and limiting, it is hypothesised that deep learning based models can make this process more efficient. However, although deep learning-based methods have achieved remarkable results in the field of animal recognition, some problems can be encountered that limit their practical application. Generally, these networks are tested on similar images taken from the dataset on which they are trained. Although the test performance of these models is high, they may perform poorly on images with different features. For this reason, in the present study on the YOLOv5 model, a number of effective preprocesses are included for the model's ability to identify and track sheep from sheep images with different traits from the training data. In addition, some adaptive adjustments were made to the YOLOv5 model to increase its effectiveness in practical applications. According to the experimental results of this study, in which videos of 20 Norduz sheep in the scale and arena tests were used, the YOLOv5l model trained on the scale test reached a mAP value of 0.99. Although the model performed the task of identifying and tracking the sheep in the scale test, it was observed that it could not perform the task of identifying and tracking the sheep in the arena test. Therefore, YOLOV5l (Model II), which was retrained on the scale images segmented from the background, gained the ability to identify and track the sheep in the arena test with some various pre-tuning. The findings of the study indicate the potential of deep learning-based models to improve the effectiveness of animal identification and tracking procedures in precision livestock farming. At the same time, the developmental stages outlined in this study provide a reference for the identification and tracking of sheep or alternative livestock in real-life situations.
Sheepfearnet: Sheep Fear Test Behaviors Classification Approach From Video Data Based on Optical Flow and Convolutional Neural Networks
(Elsevier Sci Ltd, 2023) Bati, Cafer Tayyar; Ser, Gazel
Determining the temperament related traits of sheep, such as the coping style with various stress factors such as people, a new environment and social isolation, is essential in terms of improving animal welfare and increasing productivity. The classification of sheep according to their behavioral responses to the mentioned stress factors is evaluated by objective or subjective methods by expert observers. However, visual examinations that rely on human observation are more likely to make mistakes and are time consuming. Therefore, it is important to make this process faster, easier and more reliable. The phenotypic and genetic heritability of temperament traits in sheep are examined using behavioral tests such as arena and isolation box. The spatial features of the temper-ament classes in these tests are generally similar. At the same time, since the behavior traits are composed of time series, defining or classifying these features with image-based approaches can present challenges. In this study, we propose a video-based approach to overcome this challenge, using videos of behavioral traits obtained from fear tests. In this approach, we used a combination of optical flow for capturing temporal features and con-volutional neural networks for capturing spatial features. The experimental results show that, balanced datasets in terms of the number of sheep, the BOF-VGG19 model trained with the transfer learning method is 90%, the BOF-CovnLSTM model using ConvLSTM networks is 95%, and the BOF-CNN model using CNNs is 100%, were determined as the optical flow models that classify fear test behavior traits the most successfully. The success rate of UNB-CNN and B-CNN models trained on raw images was 70%. As a result, we obtained successful results in classifying behavioral traits in models trained with optical flow pre-processed data sets balanced in terms of sheep numbers. At the same time, using a combination of optical flow and convolutional neural networks in videos where spatial features between temperament classes are similar enhanced the classification accuracy of fear behavior traits by capturing temporal features.
The Use of Different Machine Learning Algorithms in the Determination and Classification of Fear Test Behaviors in Sheep
(2022) Bati, Cafer Tayyar; Ser, Gazel
Bu tez çalışmasında, koyunlarda mizacın sınıflandırılması amacıyla korku testi (arena testi) davranış özelliklerinin yer aldığı video kayıtları kullanılarak, makine öğrenmesi ve derin öğrenme algoritmalarından yararlanılmıştır. Bu amaçla, 3-4 yaşlı 100 baş Norduz koyununa ait arena test davranış özelliklerini içeren video görüntüleri kullanılmıştır. Çalışmanın ilk aşamasında, her hayvanın arena davranış özelliklerinin yer aldığı veri seti, dört farklı denetimsiz makine öğrenmesi algoritması kullanılarak, hayvanlar mizaç özelliklerine göre (aktif ve pasif) kümelenmiştir. İkinci aşamada, etiketlenmiş veri, altı farklı denetimli makine öğrenmesinde değerlendirilmiştir. Bu aşamada, sınıflandırma tahmin performanslarına göre en başarılı denetimsiz öğrenme algoritması belirlenmiş ve video görüntülerinin etiketlenmesi gerçekleştirilmiştir. Üçüncü aşamada, etiketlenen video görüntülerinden üç farklı veri seti oluşturulmuştur. Bu veri setleri, farklı derin öğrenme modelleri kullanılarak sınıflandırma performansları elde edilmiştir. Son aşamada ise en iyi model kombinasyonu belirlenerek, gerçek zamanlı görüntülemede kullanılmıştır. Davranış özellikleri veri setini kümelemede k-ortalamalar yöntemi ve arena davranış özelliklerini sınıflandırmada ise destek vektör makineleri ve Naive Bayes yöntemleri oldukça başarılıdır. Video görüntüleri kullanılarak yapılan üçüncü aşama sonuçlarında ise dengeli ve optik akışa sahip üçüncü veri setiyle eğitilen modellerden (model 3.1, 4.2 ve 5.3'den sırasıyla %100, %95 ve %90) başarılı performanslar elde edilmiştir. Sonuç olarak, video kayıtlarından elde edilen davranış özelliklerinin sınıflandırılmasında, optik akış yönteminin kullanımı başarılı performansların elde edilmesini sağlamıştır. Bu çalışmayla hayvan davranış özelliklerinin sınıflandırılmasında geliştirilen makine öğrenimi modellerinin, hassas hayvancılık uygulamalarına katkı sağlayacağı söylenebilir.