Browsing by Author "Dundar, Mehmet Murat"

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    Automated image classification via unsupervised feature learning by K-means
    (2015-07-09) Karimy Dehkordy, Hossein; Dundar, Mehmet Murat; Song, Fengguang; Xia, Yuni
    Research on image classification has grown rapidly in the field of machine learning. Many methods have already been implemented for image classification. Among all these methods, best results have been reported by neural network-based techniques. One of the most important steps in automated image classification is feature extraction. Feature extraction includes two parts: feature construction and feature selection. Many methods for feature extraction exist, but the best ones are related to deep-learning approaches such as network-in-network or deep convolutional network algorithms. Deep learning tries to focus on the level of abstraction and find higher levels of abstraction from the previous level by having multiple layers of hidden layers. The two main problems with using deep-learning approaches are the speed and the number of parameters that should be configured. Small changes or poor selection of parameters can alter the results completely or even make them worse. Tuning these parameters is usually impossible for normal users who do not have super computers because one should run the algorithm and try to tune the parameters according to the results obtained. Thus, this process can be very time consuming. This thesis attempts to address the speed and configuration issues found with traditional deep-network approaches. Some of the traditional methods of unsupervised learning are used to build an automated image-classification approach that takes less time both to configure and to run.
  • Thumbnail Image
    Item
    A nonparametric Bayesian perspective for machine learning in partially-observed settings
    (2014-07-31) Akova, Ferit; Dundar, Mehmet Murat; Qi, Yuan Alan
    Robustness and generalizability of supervised learning algorithms depend on the quality of the labeled data set in representing the real-life problem. In many real-world domains, however, we may not have full knowledge of the underlying data-generating mechanism, which may even have an evolving nature introducing new classes continually. This constitutes a partially-observed setting, where it would be impractical to obtain a labeled data set exhaustively defined by a fixed set of classes. Traditional supervised learning algorithms, assuming an exhaustive training library, would misclassify a future sample of an unobserved class with probability one, leading to an ill-defined classification problem. Our goal is to address situations where such assumption is violated by a non-exhaustive training library, which is a very realistic yet an overlooked issue in supervised learning. In this dissertation we pursue a new direction for supervised learning by defining self-adjusting models to relax the fixed model assumption imposed on classes and their distributions. We let the model adapt itself to the prospective data by dynamically adding new classes/components as data demand, which in turn gradually make the model more representative of the entire population. In this framework, we first employ suitably chosen nonparametric priors to model class distributions for observed as well as unobserved classes and then, utilize new inference methods to classify samples from observed classes and discover/model novel classes for those from unobserved classes. This thesis presents the initiating steps of an ongoing effort to address one of the most overlooked bottlenecks in supervised learning and indicates the potential for taking new perspectives in some of the most heavily studied areas of machine learning: novelty detection, online class discovery and semi-supervised learning.
  • Thumbnail Image
    Item
    A novel machine learning model for class III surgery decision
    (Springer, 2022-08) Lee, Hunter; Ahmad, Sunna; Frazier, Michael; Dundar, Mehmet Murat; Turkkahraman, Hakan; Orthodontics and Oral Facial Genetics, School of Dentistry
    Purpose The primary purpose of this study was to develop a new machine learning model for the surgery/non-surgery decision in class III patients and evaluate the validity and reliability of this model. Methods The sample consisted of 196 skeletal class III patients. All the cases were allocated randomly, 136 to the training set and the remaining 60 to the test set. Using the test set, the success rate of the artificial neural network model was estimated, along with a 95% confidence interval. To predict surgical cases, we trained a binary classifier using two different methods: random forest (RF) and logistic regression (LR). Results Both the RF and the LR model showed high separability when classifying each patient for surgical or non-surgical treatment. RF achieved an area under the curve (AUC) of 0.9395 on the test set. 95% confidence intervals were computed by bootstrap sampling as lower bound = 0.7908 and higher bound = 0.9799. On the other hand, LR achieved an AUC of 0.937 on the test set. 95% confidence intervals were computed by bootstrap sampling as lower bound = 0.8467 and higher bound = 0.9812. Conclusions RF and LR machine learning models can be used to generate accurate and reliable algorithms to successfully classify patients up to 90%. The features selected by the algorithms coincide with the clinical features that we as clinicians weigh heavily when determining a treatment plan. This study further supports that overjet, Wits appraisal, lower incisor angulation, and Holdaway H angle can be used as strong predictors in assessing a patient’s surgical needs.
  • Thumbnail Image
    Item
    Prediction of the Post-Pubertal Mandibular Length and Y Axis of Growth by Using Various Machine Learning Techniques: A Retrospective Longitudinal Study
    (MDPI, 2023-04-26) Wood, Tyler; Anigbo, Justina O.; Eckert, George; Stewart, Kelton T.; Dundar, Mehmet Murat; Turkkahraman, Hakan; Orthodontics and Oral Facial Genetics, School of Dentistry
    The aim was to predict the post-pubertal mandibular length and Y axis of growth in males by using various machine learning (ML) techniques. Cephalometric data obtained from 163 males with Class I Angle malocclusion, were used to train various ML algorithms. Analysis of variances (ANOVA) was used to compare the differences between predicted and actual measurements among methods and between time points. All the algorithms revealed an accuracy range from 95.80% to 97.64% while predicting post-pubertal mandibular length. When predicting the Y axis of growth, accuracies ranged from 96.60% to 98.34%. There was no significant interaction between methods and time points used for predicting the mandibular length (p = 0.235) and Y axis of growth (p = 0.549). All tested ML algorithms accurately predicted the post-pubertal mandibular length and Y axis of growth. The best predictors for the mandibular length were mandibular and maxillary lengths, and lower face height, while they were Y axis of growth, lower face height, and mandibular plane angle for the post-pubertal Y axis of growth. No significant difference was found among the accuracies of the techniques, except the least squares method had a significantly larger error than all others in predicting the Y axis of growth.