Theses supervised by Prof. Dr. Mehmet Fatih Akay
11 theses · Çukurova University
Predicting maximal oxygen uptake using deep learning
Maximal oxygen uptake, or VO2max, is an external parameter that is affected by things like how many red blood cells the body has, how adapted the muscles are to distance running, and how much blood the heart can pump. It is measured as milliliters of oxygen used in one minute per kilogram of body weight. In a laboratory, it is calculated by measuring the volume (V) of oxygen (O2) that the body consumes while running on a treadmill which is the most accurate way. However, because of the serious drawbacks of direct measurement, a lot of studies have been conducted using machine learning methods to predict VO2max. The purpose of this study is to build new VO2max prediction models using deep learning (DL). The dataset has been split into training and test data using 70-30%, 80-20% split ratio, and 10-fold cross-validation. For comparison purposes, VO2max prediction models based on Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and Single Decision Tree (SDT) have also been developed. The performance of the prediction models has been evaluated using Standard Error of Estimate (SEE) and Multiple Correlation Coefficient (R). As a conclusion, DL can be used safely in VO2max prediction domain.
Development of a new software for resting metabolic rate prediction using machine learning methods
Resting metabolic rate(RMR) indicates the number of calories that are needed to carry out basic functions in mammals like blood circulation, brain functions, breathing, fuel ventilation, temperature regulation, etc. at complete rest. Accurate prediction of RMR plays a critical role to detect an individual's establish daily calorie needs, risk of heart disease and stroke, hypertension, diabetes, and internal age so nutritionists calculate RMR of patients within the specified period and able to prepare diet lists for them to help their wellness goals. In this thesis, it was objected to develop a novel web-based application that can predict the individual's RMR using different machine learning methods. The application has been developed using DTREG predictive modeling software library, Visual Studio, and C# programming language. Three different machine learning methods which are General Regression Neural Network (GRNN), Multi-Layer Perceptron (MLP), Support Vector Machines (SVM) have been integrated into the software. Different prediction models have been assessed according to their Root Mean Square Error (RMSE) metrics. As a result, it has been proven that this software can be used for RMR prediction, producing acceptable error rates under certain circumstances.
Prediction of electricity market clearing price using machine learning and deep learning
The market clearing price is the equilibrium monetary value of a traded asset or good and is an important metric in the calculation of electricity prices. In our country, when the electricity price is determined by the companies and in the price changes, market clearing prices are calculated. In this study, forecasting models consisting of 24 forecasts were created in order to make one day forecasts using hourly market clearing price data. The main objective of this thesis is to estimate the market clearing price of electricity using Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM), which are machine learning and deep learning methods and to create appropriate models. The performance of the predicted models was evaluated by calculating the Mean Absolute Percent Error (MAPE) value. The results generally show that LSTM (Long Short Term Memory) and CNN (Convolutional Neural Network) based models perform better than other methods. Keywords: Machine and Deep Learning, Forecast, Market Clearing Price
Predicting the performance of cross-country skiers using maching learning methods
The purpose of this thesis is to develop new regular and feature selection-based models for predicting the racing times of cross-country skiers by using machine learning and feature selection methods. Particularly, six popular machine learning methods including Optimized-General Regression Neural Network (OPGRNN), General Regression Neural Network (GRNN), Support Vector Machine (SVM), Multilayer Perceptron (MLP), Radial Basis Function Neural Network (RBFNN), and Single Decision Tree (SDT) have been used, whereas Relief-F has been employed as the feature selector. Several models have been developed to predict the racing time of cross-country skiers using physiological data along with a rich set of survey-based data. By performing 10-fold cross-validation, the prediction errors of the models have been calculated using root mean square error (RMSE). The results emphasize that OPGRNN-based prediction models show superior perfor¬mance and can be categorized as a feasible tool to predict the racing time of cross-country skiers. Furthermore, significant advantages such as the non-exercise-based usage and the applicability to a broader range of cross-country skiers make the prediction models proposed in this study easy-to-use and more valuable. Key Words: Machine learning, racing time, cross-country skiers, prediction
Anomaly detection in network traffic using machine learning
A primary thematic of this study is centered on detecting anomalies and measuring the device health for Central Processing Unit (CPU), memory utilization, and allocation; for Key Performance Indicator (KPI) dataset which assembled throw twenty-one-day, by improving models using machine learning (ML) methods; namely, Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM), with Auto Encoders (AE), One-Class Support Vector Machine (Oc-SVM), also k-Nearest Neighbors (k-NN). The accuracy of all methods was measured by using a confusion matrix. According to the observed results, the deep learning methods yield great performance results compared to classification methods for all models. In general, CNN/AE and LSTM/AE models show higher accuracy than the other methods. The ranking of models from best to worst based on accuracy in the confusion matrix are; CNN/AE, LSTM/AE, as for the deep learning models, while for classification models the favorable order for the methods are; k-NN, and Oc-SVM.
Efficient scaling with machine learning on cloud environment
Scaling in cloud environments can significantly affect the cost and efficiency of a system. In order to properly plan capacity, calculations are often made based on scaling size. Applications typically reserve resources such as cores, memory, and network in order to maintain high quality of service (QoS). When resource limits are approached, the application is scaled by running multiple copies in order to ensure system reliability under increasing traffic. This study aims to more efficiently reserve resources in order to maintain high QoS through the use of machine learning. Traditional methods reserve resources for an application and scale the application when resource usage reaches a certain threshold. However, the method implemented in this study estimates incoming traffic and scales applications based on this estimation, resulting in more efficient performance compared to using a fixed threshold value. The results of the study showed that the scaling system created based on machine learning performs more efficiently in terms of cost and resource allocation compared to the currently used static scaling methods. Keywords: Machine Learning, Cloud Computing, Scaling, Kubernetes, Resource Management, Cloud Cost
Bitcoin price prediction with machine learning
The most well-known cryptocurrency is Bitcoin, as it was the first cryptocurrency. The approximate value of the cryptocurrency market capitalization, of which 39% is Bitcoin, is 807 billion dollars (December 2022). It has become an important research topic due to the difficulty in predicting the price of Bitcoin due to the extreme volatility of the price, and there are many studies in the literature. For investors, this high volatility means high profits and risks. This thesis aims estimation results close to the actual price and reduces the risks for the investors, using machine learning methods, different data sets, and optimization techniques. The methods used: Multilayer Perceptron (MLP), Support Vector Machines (SVM), Generalized Regression Neural Network (GRNN), Recurrent Neural Network (RNN), Long-Short-Term Memory (LSTM), Gated Repetitive Unit (GRU), and Convolutional Neural Network (CNN). The thesis includes experiments with each machine learning model with the combinations of Bitcoin, gold, crude oil, natural gas, Ethereum, and dollar-euro parity. Hyper-parameter optimization methods such as Bayesian optimization (BO), random search, grid search, and Hparam parameters are examined. Models with BO achieved better results than others. This thesis proposes a new model for Bitcoin price prediction that effectively reduces prediction error. This new BO model with Gradient Incremental Regression Trees (GBRT), Gaussian Process (GP), Random Forest (RF), and Extra Trees (ET) was applied to optimizers and corresponding surrogate functions. In addition, to increase the comparability of the results with the other paper, it was evaluated with four different performance metrics: root square mean error (RMSE), mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In general, among the seven algorithms, predictions using only the closing price of Bitcoin yielded better results. In addition, we obtained very close results in the estimates made by adding Ethereum, crude oil, and natural gas to the data set. Better results were obtained with LSTM, CNN, and GRU, respectively, than with the other methods. The experimental optimization results indicated that hparam, grid search, and random search achieved the worst results in all four error metrics. BO-GP with hybrid LSTM-GRU outperformed all methods in this thesis and the examined literature for the value of MAE=0.002302, MAPE=0.005497, MSE=0.000015, and RMSE=0.003269.
Yapay zeka ve istatistiksel yöntemler kullanarak yeni elektrik sistemi marjinal fiyat tahmin modellerinin geliştirilmesi
Elektrik sektöründe, SMP, enerji talebi ve arzının dengelenmesinde kritik bir belirleyici rol oynamaktadır. SMP değerleri, enerji talebindeki değişiklikler, arz kaynaklarındaki varyasyonlar, politika düzenlemeleri ve çevresel koşullar gibi çeşitli karmaşık faktörlere bağlı olarak sürekli dalgalanmalar göstermektedir. Bu dalgalanmalar, SMP'yi enerji üreticileri, dağıtıcıları ve tüketicileri için enerji piyasasındaki dinamik değişkenleri yansıtan hayati bir metrik haline getirmektedir. Enerji kaynaklarının arz-talep dengesi, enerji ithalat ve ihracatındaki değişiklikler, politika müdahaleleri ve teknolojik gelişmeler, marjinal fiyatların belirlenmesinde kilit etkiler arasında yer almaktadır. Bu bağlamda, SMP, elektrik sektöründeki karar vericiler için stratejik bir öneme sahiptir ve enerji şirketleri ile düzenleyicilerin enerji piyasasının karmaşıklıklarına daha iyi uyum sağlamalarına ve etkin enerji yönetimini güvence altına almalarına yönelik önemli bilgiler sunmaktadır. Bu çalışmanın temel amacı, Türk elektrik piyasası için hiperparametrik esneklik temelinde bir arayüzü içeren SMP tahmin modelleri geliştirmektir. Bu amaca ulaşmak için, yapay zeka teknikleri ve ileri istatistiksel yöntemler kullanılarak yüksek doğrulukta SMP tahminleri elde edilmektedir. Özellikle bu tezde, SMP tahmin modellerinin performansını artırmak amacıyla mRMR ve MLFS gibi öznitelik seçimi algoritmaları uygulanmaktadır. Bu çalışmada kullanılan veri seti, 1 Ocak 2021 ile 14 Eylül 2023 tarihleri arasını kapsamakta olup EPİAŞ tarafından sağlanmıştır. Sonuçlar, ELM hem de XGBoost tabanlı modellerin karşılaştırılabilir derecede güçlü bir performans sergilediğini ve SMP tahminlerinde etkili olma potansiyellerini vurgulamaktadır. Buna karşılık, MC yöntemi genellikle kötü sonuçlar üretmekte olup, SMP tahminleri için sınırlı bir etkinlik göstermektedir. Anahtar Kelimeler: Elektrik Enerjisi Sektörü, Makine Öğrenmesi, Sistem Marjinal Fiyatı
Forecasting arrivals to a call centre using machine learning and deep learning
With the increasing complexity of call center operations, accurate call arrival prediction has become a crucial research area, attracting significant attention from both academia and industry. Forecasting call arrivals is essential for effective resource allocation, staffing decisions, and service level planning, ultimately improving operational efficiency and customer satisfaction. This study investigates the effectiveness of combining Bayesian optimization (BO) with feature selection to enhance call arrival prediction accuracy. We analyzed various machine learning (ML) and deep learning (DL) models, including Random Forest (RF), Multi-layer Perceptron (MLP), Support Vector Machine (SVM), Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). Using real-world call center data, we employed three datasets with daily, hourly, and half-hourly observations to predict call volume and Average Handling Time (AHT). Autocorrelation functions revealed patterns in these datasets. Both multivariate and univariate prediction approaches were evaluated. Using the Mean Absolute Error (MAE) metric and Root Mean Squared Error metric (RMSE), we demonstrated that optimized models with selected features consistently outperformed baseline models and optimized models with all features. Specifically, DL models, notably CNN and LSTM, showed robust responsiveness to combined BO and feature selection. ML models, especially MLP and RF, had strong baseline performances and also benefited from our approach. Depending on the dataset, performance gains from our approach ranged from 14% to 88% for ML models and 92% to 97% for DL models for both call volume and AHT predictions for multivariate models. For univariate models with only BO, percentage improvements ranged from 20% to 71% for ML models and 50% to 96.6% for DL models. These results highlight the effectiveness of combining feature selection with BO for more accurate predictions of both call volume and AHT. Keywords: Call Arrivals forecasting, machine learning, Hyperparameter Optimization, Feature Selection
E-ticaret sektörü için yapay zeka tabanli satiş i̇ptal/i̇ade tahminleme yazilimi
In today's rapidly evolving digital landscape, e-commerce has moved far beyond serving as a simple sales channel and has become a core component of strategic business planning. This study aims to develop machine learning based classification models for order cancellation and return prediction. These models were built using a real-world dataset of 11,000 transaction records collected from an e-commerce company between May 2020 and August 2024. 10,000 records were used for training the models, while the remaining 1,000 were reserved for independent testing. The training dataset was balanced across three classes: completed, canceled, and returned orders. Two distinct modeling approaches were adopted. The first approach utilized a multi-class classification framework to predict all three outcomes in a single model. The second approach employed separate binary classification models for cancellations and returns, allowing for more targeted predictions. To construct these models, five machine learning algorithms were applied: Logistic Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Deep Neural Network (DNN). To enhance model performance, four feature selection strategies were evaluated: minimum Redundancy Maximum Relevance (mRMR), Relief-F, F-Classification, and a Hybrid method combining the three. Among these, Relief-F and Hybrid approach yielded the most robust results, particularly when used in conjunction with RF and XGBoost.The results indicated that the binary classification approach produced superior predictive accuracy and sensitivity compared to the multi-class approach. SVM models demonstrated strong performance in binary classification scenarios, whereas LR and DNN models underperformed. Overall, the integration of advanced feature selection techniques with RF and XGBoost significantly improved forecasting capabilities, offering e-commerce businesses a valuable tool for optimizing operational planning and decision-making processes.
Breast cancer classification using effective machine learning techniques
Breast cancer remains a leading cause of death among women worldwide, underscoring the urgent need for practical diagnostic tools. This work presents an advanced machine learning algorithm designed to enhance the classification accuracy of breast cancer. The system integrates a deep multi-layer perceptron (Deep MLP) for feature extraction, a feature-fused autoencoder for efficient dimensional reduction, and a weight-tuned decision-tree classifier optimized by cross-validation and square weight adjustment. The Wisconsin breast cancer dataset is utilized to test the results of the method rigorously using k-fold cross-validation. Optimizing performance has been done under different hyperparameters, namely, the number of hidden units, dropout rate, batch size, as well as test-train percentages. The performance of the model was evaluated under all the given conditions for key Metrics. These are the Accuracy, Precision, Recall, F1-score, and area under the curve (AUC). Using the above metrics, the model was able to distinguish malignant and benign tumors. Our findings show that this approach performs better than traditional classification methods, leading to accurate and robust results across several data partitions. This research contributes to a new framework pertaining to deep learning, Auto-encoder, and decision tree, which clearly shows that this framework has a very high probability of impacting breast cancer diagnosis while providing usefulness to physicians. Furthermore, the use of the METABRIC dataset ,which is substantially larger and more diverse than classical breast cancer datasets ,provided a robust evaluation environment for the proposed hybrid model. Its rich clinical and genomic features enabled deeper validation of the model's generalization capability and demonstrated its effectiveness in handling complex, real-world data.