Theses supervised by Doç. Dr. Mehmet Fatih Akay
15 theses · Çukurova University
Prediction of upper body power and maximal oxygen uptake of cross-country skiers using different regression methods
Upper body power (UBP) and maximal oxygen uptake (VO2max) are the two most important determinants of cross-country ski race performance. Although numerous studies exist to measure UBP of cross-country skiers, to date, no study has attempted to predict UBP of cross-country skiers. The purpose of this thesis is to develop prediction models for estimating 10-second UBP (UBP10), 60-second UBP (UBP60) and VO2max of cross-country skiers using different regression methods, namely support vector machines using the radial-basis function (SVM-RBF), linear SVM (SVM-Linear), multi layer perceptron (MLP) and multiple linear regression (MLP). Several UBP and VO2max prediction models have been developed using different data sets and combination of the predictor variables such as protocol, age, gender, height, weight, body mass index (BMI), heart rate (HR), heart rate at lactate threshold (HRLT) and exercise time. By using 10-fold cross-validation on the data sets, the performance of the models has been evaluated by calculating their standard error of estimates (SEE's) and multiple correlation coefficients (R's). The results show that SVM-RBF-based UBP and VO2max prediction models perform better (i.e. yield lower SEE's and higher R's) than the prediction models developed by other regression methods.
Predicton of maximal oxygen uptake using machine learning methods combined with feature selection
Maximal oxygen uptake (VO2max) refers to the maximum amount of oxygen that an individual can utilize during intense or maximal exercise. Although numerous studies exist to predict VO2max, to date, no study has attempted to apply machine learning methods combined with feature selection algorithms to identify the discriminative features for prediction of VO2max. The purpose of this thesis is to develop new VO2max prediction models using Support Vector Machines (SVM) and Multilayer Perceptron (MLP) combined with feature selection algorithms. Two feature selection algorithms, Relief-F and Correlation-based Feature Selection (CFS), have been considered. By applying Relief-F on the full data sets, the ranking of the features have been obtained. Dimensionality of data sets is reduced by removing the feature with the lowest score at a time before being passed on to the regression method. CFS has been used separately on the full data sets to find out the best subset of features. Using 10-fold cross validation on four different data sets, the performance of prediction models has been evaluated by calculating their multiple correlation coefficients (R's) and standart error of estimates (SEE's). The results show that SVM-based VO2max prediction models perform better (i.e. yield lower SEE's and higher R's) than the prediction models developed by other regression methods.
Prediction of maximum oxygen uptake (VO2max) with submaximal and questionnaire variables using different regression methods
Maximal oxygen uptake (VO2max) refers to the maximum amount of oxygen that an individual can utilize during intense or maximal exercise. The purpose of this thesis is to develop accurate VO2max prediction models using submaximal and questionnaire variables. Regression methods such as Support Vector Machines (SVM), Multilayer Perceptron (MLP) and Multiple Linear Regression (MLR) have been used for developing VO2max prediction models. The performance of prediction models has been evaluated by calculating their multiple correlation coefficients (R's) and standard error of estimates (SEE's). The results show that the accuracy of VO2max prediction models based on submaximal and standard non-exercise variables could be significantly improved by including questionnaire variables in prediction models. The results of SVM models have been also compared with the ones obtained by MLP and MLR and it turned out that SVM-based VO2max prediction models perform better (i.e. yield lower SEE's and higher R's) than the prediction models developed by other regression methods.
Developing VO2max prediction models from non-exercise, submaximal exercise and hybrid tests using machine learning methods
Maximum oxygen uptake (VO2max) is defined as the maximum ability to transport and consume oxygen during strenuous endurance exercise and is considered the single best measure of cardiorespiratory fitness (CRF). The purpose of this thesis is to develop VO2max prediction models by using non-exercise, submaximal and hybrid variables by using Support Vector Machines (SVM), Multi-layer Feed-forward Artificial Neural Networks (MFANN) and Multiple Linear Regression (MLR) on different data sets. Using 10-fold cross validation on four different data sets, the performance of prediction models has been evaluated by calculating their multiple correlation coefficients (R's) and standard error of estimates (SEE's). The results show that SVM-based VO2max prediction models perform better (i.e. yield lower SEE's and higher R's) than the prediction models developed by MFANN and MLR. We also propose a new approach based on the elimination of irrelevant samples during the training phase for improving the performance of SVM and MFANN models for prediction of VO2max. The performance of the proposed approach has been compared with the two widely used outlier detection algorithms. The results show that the improved SVM-based and MFANN-based VO2max prediction models yield noticeable decrements in error rates compared to that of regular and outlier-based SVM and MFANN VO2max prediction models.
Predicting the admission decision acandidate to the School of Physical Education and Sport at Çukurova University by using different machine learning algorithms
The purpose of this thesis is to develop new prediction to estimate the admission decision of a candidate to the School of Physical Education and Sports at Cukurova University ones he knows his skores from the physical ability test. The classifiers used to evaluate the performance of the prediction models include Support Vector Machine (SVM), Multilayer Perspectron (MLP), Logistic Regression(LR), Radian Basis Function (RBF), Network, Single Desicion Tree(SDT),and K-Means Clustering. Experiments have been condusted on two datesets that include real test result of the candidates who applied to the School in 2006 and 2007, respectively. For model testing and validation ,5-fold and 10-fold cross validation as well as several diffrent percent splits of training/testing date have been used. The performance of the classifiers on the datasets has been evaluated by calculating the classification accuracy and several other performance metrics. The results show that classification accuracy of the SVM classifier using 10-fold cross validation achieves the highest accuracy with 97.90% and 91.45% for the 2006 and 2007 datasets respectively. The rank ing among the six classifiers in terms of achieved classification accuracy has been determined as SVM, LR, MLP, RBF, SDT AND K-Means Clustering. Key Words: Machine Learning, Physical Ability Test, Prediction,Desicion Support Systems.
Predicting the admission decision of a participant to the school of physical education and sports at Cukurova University by using different machine learning methods combined with feature selection
The purpose of this thesis is to develop new hybrid admission decision prediction models by using different machine learning methods including Support Vector Machines (SVM), Multilayer Perceptron (MLP), Radial Basis Function (RBF) Network, TreeBoost (TB) and K-Means Clustering (KMC) combined with feature selection algorithms to investigate the effect of the predictor variables on the admission decision of a candidate to the School of Physical Education and Sports at Cukurova University. Three feature selection algorithms including Relief-F, F-Score and Correlation-based Feature Selection (CFS) have been considered. Experiments have been conducted on the datasets, which contain data of participants who applied to the School in 2006 and 2007. The datasets have been randomly split into training and test sets using 10-fold cross validation as well as different percentage ratios. The performance of the prediction models for the datasets has been assessed using classification accuracy, specificity, sensitivity, positive predictive value (PPV) and negative predictive value (NPV). The results show that a decrease in the number of predictor variables in the prediction models usually leads to a parallel decrease in classification accuracy.
Predicton of upper body power of cross-country skiers using machine learning methods combined with feature selection algorithms
Upper body power (UBP) is one of the most important factors affecting the performance of cross-country skiers during races. Although some initial studies have already attempted to predict UBP, until now, no study has attempted to apply machine learning methods combined with various feature selection algorithms to identify the discriminative features for prediction of UBP. The purpose of this study is to develop new prediction models for predicting the 10-second UBP (UBP10) and 60-second UBP (UBP60) of cross-country skiers by using General Regression Neural Networks (GRNN), Radial-Basis Function Network (RBF), Multilayer Perceptron (MLP), Support Vector Machine (SVM), Single Decision Tree (SDT) and Tree Boost (TB) along with the Relief-F feature selection algorithm, minimum redundancy maximum relevance (mRMR) feature selection algorithm and the Correlation-based Feature Subset Selection (CFS). Several models have been developed to predict UBP10 and UBP60 of cross-country skiers using two datasets. 10-fold cross validation has been performed for model testing. The efficiency of the prediction models has been calculated with their multiple correlation coefficients (R's), standard error of estimates (SEE's) and mean absolute percentage errors (MAPE's). The results emphasize that GRNN-based prediction models show higher performance than the other regression methods. Also, using less number of predictor variables than the full set of predictor variables can be useful for prediction of UBP10 and UBP60 with comparable error rates.
Development of new models using machine learning methods combined with different time lags for network traffic forecasting
The purpose of this thesis is to forecast the amount of network traffic in Transmission Control Protocol/Internet Protocol (TCP/IP) -based networks by using different time lags and various machine learning methods including Support Vector Machines (SVM), Multilayer Perceptron (MLP), Radial Basis Function (RBF) Neural Network, M5P (a decision tree with linear regression functions at the nodes), Random Forest (RF), Random Tree (RT), and Reduced Error Prunning Error (REPTree), and statistical regression methods including Multiple Linear Regression (MLR) and Holt-Winters and compare the performance of statistical and machine learning methods. Two different Internet Service Providers' (ISPs) traffic data have been utilized to build traffic forecasting models. The first 66% of the data sets has been utilized as training sets and the rest has been used as test sets. The performance of the forecasting models for the data sets has been assessed using Mean Absulote Percentage Error (MAPE). The results show that SVM and M5P based models usually perform better than the ones obtained by the other methods.
Development of new hybrid models for prediction of VO2MAX using machine learning methods combined with feature selection algorithms
The purpose of this thesis is twofold. The first purpose is to develop new hybrid feature selection-based maximal oxygen uptake (VO2max) prediction models using for the first time the double and triple combinations of maximal, submaximal and questionnaire variables. Several machine learning methods including Support Vector Machine, artificial neural network-based and tree-structured methods combined individually with three feature selectors Relief-F, minimum redundancy maximum relevance (mRMR) and maximum-likelihood feature selector (MLFS) have been applied for model development. The second purpose is to design a new ensemble feature selector, which aggregates the consensus properties of Relief-F, mRMR and MLFS to produce more robust decisions about the set of relevantly identified VO2max predictors and to create more accurate prediction models. Using 10-fold cross validation on three different datasets, the performance of prediction models has been evaluated by calculating their multiple correlation coefficients (R's) and root mean squared errors (RMSE's). The results show that compared with the results of the other regular feature selection-based models in literature, the reported values of R and RMSE of the hybrid models in this thesis are considerably more accurate. Furthermore, prediction models based on the proposed ensemble feature selector outperform the models created by individually using the Relief-F, mRMR or MLFS, achieving similar or ideally up to 12.46% lower error rates on the average.
Voice and data traffic modeling and prediction for a third generation mobile network using machine learning methods
The purpose of the thesis is to derive models for traffic characteristics of a 3G network which is commercially deployed in Turkey and predict voice and data traffic by using various machine learning methods. The machine learning methods which were employed are Support Vector Machines (SVM), Multilayer Perceptron (MLP), Random Forest (RF) and Radial Basis Function Neural Network (RBF). Additionally, the Holt-Winters method has been applied to develop prediction models as a statistical method. Four different type of UMTS network traffic data have been utilized in order to build traffic prediction models. The performance of the forecasting models for the data sets has been assessed using Mean Absolute Percentage Error (MAPE). Finally, the performance of statistical and machine learning regression methods have been compared and the results show that SVM and Holt-Winters based models usually perform better than the ones obtained by the other methods.
Prediction of Hamstring and Quadriceps muscle strength of athletes using machine learning methods
The success of athletes is closely related to the muscles of the thigh, namely the hamstring and quadriceps. Among the different techniques used for measuring the hamstring and quadriceps muscle strength, the use of isokinetic equipments is the most accurate. However, their utilization is associated with several difficulties and limitations. The aim of this study is to build new and more comprehensive models for predicting the hamstring and quadriceps muscle strength of athletes using four main machine learning methods, namely Support Vector Machine (SVM), Multilayer Perceptron Neural Network (MLP), Radial Basis Function Neural Network (RBFNN) and Single Decision Tree (SDT). The root mean square errors (RMSE's) and the multiple correlation coefficients (R's) have been used for computing the prediction errors. On the basis of the results obtained, it has been proved that machine learning methods especially SVM can be used for the hamstring and quadriceps muscle strength prediction with an acceptable accuracy
Prediction of maximum muscular endurance time involving four stabilization exercises assessments using machine learning methods
The muscular endurance time is viewed an important component influencing the performance of athletes in various sport branches, such as cycling, rowing, cross-country skiing, swimming and running. Due to several drawbacks of direct measurement, researchers need alternative ways to determine maximum muscular endurance time. The aim of this thesis is to build new models for predicting the maximum endurance time using demographic variables (age, gender, height, weight and body mass index), submaximal data (rating of perceived exertion) and machine learning methods including support vector machines (SVM), multilayer perceptron (MLP), generalized regression neural network (GRNN), radial basis function (RBF) and single decision tree (SDT). The root mean square error (RMSE) and multiple correlation coefficient (R) have been used for assesing the performance of prediction models. The results suggest that SVM is a viable method for maximum endurance time prediction with an acceptable accuracy.
Development of a new software for network traffic prediction and forecasting using time series multilayer perceptron and feedback delays
Accurate prediction of Internet network traffic plays an important role both in academical literature and the Internet network industry. World-known corporations like Internet Service Providers (ISP), hosting companies and all types of web sites can benefit from the prediction of the amount of network data usage by arranging their business plans according to the customers' needs and preferences. In this thesis, it was aimed to develop a new software that can predict the Internet data traffic using Multilayer Perceptron (MLP) combined with Time Series Analysis. The software has been developed using MATLAB programming language. Two different prediction modes have been integrated into the software, including 'Predict' (open-loop prediction) and 'Forecast' (closed-loop prediction). The prediction models have been evaluated with respect to their Mean Absolute Percentage Error (MAPE) values. As a result, it has been proven that this software can be used for network traffic prediction, producing acceptable error rates under certain circumstances.
Development of physical fitness prediction models for Turkish secondary school students using machine learning methods
Physical fitness is a necessary component for daily activities and has significant effect on our bodies. Because of this importance of physical fitness, maintenance of physical fitness is essential for health and well-being. Physical fitness is a set of attributes that are either health or skill-related, which can be measured with specific tests, and maintaining is essential for the quality of life. However, there are certain difficulties associated with the direct measurement of physical fitness such as the high cost of equipment, availability of experienced staff and the long assessment time. Since the measurement of physical activity has a key role in performing physical fitness rate, researchers need different ways to determine physical fitness. The aim of this thesis is to develop new prediction models for Turkish secondary school students by using machine learning methods including Support Vector Machines (SVM), Radial Basis Function Neural Network (RBFNN), General Regression Neural Network (GRNN), and Single Decision Tree (SDT). The performance of the SVM-based, GRNN-based, RBFNN-based and SDT-based models have been evaluated by using 10-fold cross-validation and the root means square errors (RMSEs) have been used to compute the errors of prediction. On the scope of the results, this thesis has showed the efficiency of machine-learning methods to indicate high accuracy of the physical fitness prediction.
Development of a new software for fabric defect detection and classification using image processing and machine learning methods
Quality control in the fabric industry involves a set of standards or guidelines that help guarantee a product meets certain parameters as well as customer satisfaction. Fabric defect detection (also called inspection) is a quality control process aimed at identifying and locating defects. The aim of this thesis is to build an application based on image processing and deep learning methods to automatically detect the defects on the fabric surface and classify them. Discrete Fourier transform (DFT). Normalized cross-correlations, homogeneity equalization, and Gabor filters have been used in image processing, Faster Region Proposal Networks (Faster-R CNN) has been used in classification. The mean square errors (RMSE's) has been used for computing the detecting and classification errors. Based on the results obtained, it has been proved that image processing methods especially DFT can be used for the defect detection with acceptable results.