Search

Due to the excessive temperature rise in the Central Processing Unit (CPU), computers shut down and system damage occurs over time. In this study, a new thermoelectric cooling system is designed to reduce the temperature in the CPU. The novel cooling system is designed using a thermoelectric module. It is to remove the excess heat by conduction and convection by taking advantage of the temperature difference between the thermoelectric cooler and the CPU we add to the system. Since the temperature of the thermoelectric cooler will always be lower than the CPU temperature, effective cooling will be provided. A special electronic circuit and software have been developed for the control of the cooling unit. Three different artificial intelligence models (artificial neural network, random forest, and k-nearest neighbor) were created to dynamically control the additional cooling system and their successes were compared. Artificial intelligence determines the power and fan speed of the thermoelectric cooling system. It performs this control by evaluating all parameters (different values such as CPU frequency, voltage, number of processes) instead of a specific CPU load or a specific temperature value. While the CPU temperature was 41°C at maximum load, this temperature was reduced to 31°C thanks to the designed thermoelectric cooling system. All methods provided a high classification success in education. By using the artificial neural network 97.973% classification success was achieved, whereas the classification success for random forest and k-nearest neighbour model were 97.297% and 96.306%, respectively. [ABSTRACT FROM AUTHOR]

Brain-computer interface (BCI) systems enable individuals to use a computer or assistive technologies such as a neuroprosthetic arm by translating their brain electrical activity into control commands. In this study, the use of filter-based feature selection methods for design of BCI systems is investigated. EEG recordings obtained from a BCI system designed for the control of a neuroprosthetic device are analyzed. Two feature sets were created; the first set was band power features from six main frequency bands (delta (1.0-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12-25 Hz), high-beta (25-30Hz) and gamma (30-50 Hz)) and the second set was band power features from ten frequency sub-bands (delta (1-4 Hz), theta (4-8 Hz), alpha1 (8-10 Hz), alpha2 (10-12 Hz), beta1 (12-15 Hz), beta2 (15-18 Hz), beta3 (18-25 Hz), gamma1 (30-35 Hz), gamma2 (35-40 Hz), gamma3 (40-50 Hz)). Ten filter-based feature selection methods are investigated along with linear discriminant analysis, random forests, decision tree and support vector machines algorithms. The results indicate that feature selection methods leads to a higher classification accuracy and eigen value centrality (Ecfs) and infinite feature selection (Inffs) methods have consistently provided higher accuracy rates as compared to rest of the feature selection methods. [ABSTRACT FROM AUTHOR]

The secondhand vehicle market in Türkiye has always been active. In the secondhand vehicle market, information such as brand, model, and fuel type can be analyzed, and this information can be made useful. Prices vary according to the various features of the vehicles. Machine learning techniques can predict prices and help users set prices when selling or buying vehicles. Price prediction falls under regression. Since the number of secondhand vehicles is quite high, big data systems are used. Apache Spark and its machine learning library are quite useful for this. Linear regression, decision tree regression, random forest regression, GBT regression, and isotonic regression algorithms are used for price prediction. The random forest algorithm achieved the highest success for the price prediction with 21435.09 RMSE and 0.887 R2 values. As a result of the statistical tests performed to check the significant difference between the RMSE and R2 values obtained with the random forest algorithm and other algorithms, it is concluded that the results obtained with the random forest algorithm are statistically better than other algorithms. The random forest algorithm gives better results because the algorithm performs training over multiple decision trees, its flexibility, and strong hyperparameters. [ABSTRACT FROM AUTHOR]

In this study, the effects of climate change on the energy consumption and thermal comfort of the education building were examined and the effectiveness of passive improvement scenarios based on window parameters was evaluated by machine learning and statistical analysis. The research was based on a four-stage approach based on building simulations: (i) creation and analysis of climate change scenario-modified climate datasets, (ii) climate change impact analysis on existing building, (iii) comparative analysis of improvement scenarios, and (iv) analysis of predictive models based on machine learning. An existing secondary school building in Ankara was chosen as a case study for the evaluation of the selected performance indicators. 2025 scenarios were parametrically modeled with varying window parameters. After analyzing the complete dataset generated from performance simulations with descriptive statistics, Random Forest (RF) prediction models are trained with a subset of the data for each performance indicator. For each performance indicator, the feature importance of fine-tuned RF models was calculated with 10-fold cross-validation method, and it was seen that the window SHGC value was the most critical parameter among the tested variables. Performance predictions with RF models deviate 2% on average from their actual values and imply high predictive capacity. Moreover, with the retrofit scenarios, total energy consumption showed a reduction of up to 50%, whereas a significant improvement in indoor thermal comfort was observed. The results emphasize that the right selection of window parameters in existing educational buildings has a great effect on building energy performance. The results show that machine learning can be used effectively in the adaptation processes of buildings to climate change. The method used can be extended to cover different building parameters and technologies. [ABSTRACT FROM AUTHOR]

The aim of this study is to develop a predictive model using machine learning (ML) to identify the causes of fatalities and amputations in the metal sector based on occupational accident data collected by the Turkish Ministry of Labor and Social Security (MLSS) from 2013 to 2018. It is necessary to classify and predict occupational accident reasons in detail to prevent occupational accident. Researchers have used ML algorithm to investigate correlated factors and create effective prediction models in an effort to lower occupational accidents. In this study, we used random forest (RF) which is one of the ML algorithm to predict occupational accident reasons and consequences. 10- fold cross validation model is used for model validation and it increased %4.7 of accuracy of algorithm. Accuracy of RF is found as 0.9172. We extracted important factors that affect the occupational accident reasons at metal sector using Recursive Feature Elimination (RFE) and it is found that most important factors are secondary reason of the accident, days lost and deviation. [ABSTRACT FROM AUTHOR]

This study chiefly aimed to perform flash flood susceptibility mapping by means of machine learning methods based on the records attained in the Kentucky River basin over the last two decades. To carry out analysis, one of the widely adopted practical tree-based machine learning tools, i.e., the random forest (RF) method, was utilized, while the hyperparameters (i.e., number of trees and maximum tree depth) of the RF algorithm were tuned via the particle swarm optimization (PSO) strategy. In this vein, a total of 343 flash-flooded and the same number of random (non-flash flooded) points were assigned within the Kentucky River basin boundaries. In addition, a total of 12 factors triggering flash floods have been introduced to the corresponding points and the predictions were conducted in this regard. Many performance evaluation indicators considered within the scope of this study illustrated that the hybrid PSO-RF model revealed quite accurate predictive results based on the blinded testing set; such that both flash-flooded and non-flash flooded points exist in the test set were estimated with an accuracy of 70%. In addition, one of the promising performance indicators in assessing binary classification implementations, called AUROC, was calculated as 0.79. Further analysis regarding the individual impacts of the triggering factors also highlighted that the heavy rainfall probability factor was found to be the most effective variable, followed by topography, NDVI, and curve number, respectively. On the other hand, it was concluded that the effect of the lithology on the flash flood modeling is considerably lower compared to its counterparts. Overall, the results acquired in the light of all these findings have important potential in terms of both contributing to the flood susceptibility mapping literature and guiding with respect to the measures that should be taken prior to the flash flood incidents in the corresponding region. [ABSTRACT FROM AUTHOR]

Legumes are the most basic carbohydrate, fiber and protein sources and provide essential vitamins and minerals for a healthy nutrition. As a gene center for legumes, edible legumes are the most cultivated field crops after cereals in Turkey and mainly produced ones are chickpeas, green and red lentils, beans and broad beans. Besides, as the agricultural production is very sensitive to weather conditions, it makes the current global climate change even more important because of its negative effects of on the environment, natural resources and production systems. In this study, by using a data set including meteorological factors as average humidity, average temperature, average precipitation in addition to cultivation and production values, estimations of production amounts and yields were performed for chickpeas and lentils by using Support Vector Regression (SVR), Decision Tree Regression (DTR) and Random Forest Regression (RFR) machine learning methods for 2021 and 2022. Also, accuracy and validity of the methods were evaluated by comparing obtained predictions with actual values and by considering various performance evaluation criteria. According to the obtained results by all three methods, in 2022 a 25% decrease in chickpea production and an increase of 25-65% in lentil production is predicted compared to the previous year. In addition, it can also be stated that tree-based machine learning methods yielded more successful outcomes in both predictions. [ABSTRACT FROM AUTHOR]

Objective: The aim of the study is to divide the data into appropriate subgroups using appropriate clustering methods and cluster numbers, and to analyze the changes in classification performance by performing the analyzes for the dependent variable of interest in these subgroups. Material and Methods: The data set used in the study was taken from the Kaggle database. The data set consists of 16 variables, 15 independent and 1 dependent variable, and includes 948 patient data. For the analysis, the R (ver. 4.1.2) programming language was used and the chisquare, wilcox.test, cluster, RWeka and e1071 packages in this program were used. Kmeans was used as the clustering method, and logistic regression, multilayer perceptron, decision table, decision support machine, random forest, J48, IBk, bagging and voted perceptron were used as classification methods. The dataset was tested using 10-fold cross validation option and all analyzes were repeated 1,000 time. Accuracy, F-measure, Matthews correlation coefficient, area under ROC curve and precision-recall area were used as performance criteria. Results: Considering the estimation results for patients with and without Type 2 diabetes with the models, an increase was observed in the performance of patients who were predicted to have Type 2 diabetes in the first cluster compared to the general model. An increase was observed in the performance of patients who were predicted not to have Type 2 diabetes in the second cluster compared to the general model. Conclusion: After clustering, it was observed that the specific characteristics of the patients were determined better and the models created with the cluster-based classification had better performance. [ABSTRACT FROM AUTHOR]

With digitalization, the world of crime has also become digital and the number of crimes committed over the internet is increasing day by day. Cybercriminals and attackers use secret networks on the Internet, called the Dark Web, to hide their identities and provide encrypted communication. Darknets have different and special access methods than normal internet infrastructure. All access to these networks is suspect and needs to be investigated. Because the Darknet provides encrypted communication, it is difficult to detect and classify with today's security tools. In this study, only the statistical information of packets was analyzed using machine learning approach without deciphering encrypted network traffic. CICDarknet2020 dataset was used and a detailed experimental study including K Nearest Neighbor, Logistic Regression, Random Forest, SVM, Decision Tree, Gaussian Naive Bayes, Linear Discriminatory Analysis, Gradient Boosting, Extra Tree and XGBoost algorithms was carried out for packet analysis. In experimental studies, it has been observed that the Decision Tree algorithm has the highest classification success with an accuracy rate of 93.32%. [ABSTRACT FROM AUTHOR]

The increasing urbanization and rapid economic development have led to growing housing markets. This has resulted in a rising demand for housing due to migration towards cities and the acceptance of housing as both an investment and a wealth protection tool. On the other hand, housing prices are affected by many factors, thereby studies on housing price forecasts are carried out within the scope of different models and various methods. In line with the significant importance of housing markets as one of Turkey's economic growth tools, this study includes the forecast of the classification of 4-bedroom (3+1) housing prices in the central district of Kütahya province. The dataset used includes various criteria that affect housing prices, and different machine learning classification algorithms that are utilized. As a result, it is seen that the accuracy rates of the k-Nearest Neighbor (k-NN), Decision Trees (DT) and Random Forest (RF) algorithms were above 60%. As the main performance criterion, the AUC scores yielded that the most successful classification forecast is acquired by the RF algorithm, which is followed by the k-NN. In the context of the variables and data set included in this study, it was seen that successful outcomes in the classification forecasts of 4-bedroom apartments’ housing prices in the city center of Kütahya were obtained by these two methods. [ABSTRACT FROM AUTHOR]

Diabetes is one of the common health problems with an increasing incidence worldwide. Diabetes is a chronic disease that can damage organs such as the eyes, heart, and kidneys, as well as cause mortality if not taken under control. Early diagnosis of diabetes is important in terms of preventing complications and increasing the quality of life. Machine learning techniques, which are widely used in the medical field, play the role of an intelligent decision support system that helps experts in the diagnosis of different diseases. This study includes classifications performed on the Pima Indian Diabetes dataset with six different machine learning techniques for the early diagnosis of diabetes. One of the main goals of the classifications carried out is to increase the prediction accuracy. In this study, fourteen different resampling methods were used on the dataset to increase the success of the classifiers. A total of ninety classifications were carried out without sampling and resampling for each machine learning model. The success of each classification process was reported with five different performance metrics. The highest performance was obtained with an accuracy of 96.296% in the classification using the Random Forest with the InstanceHardnessThreshold under-sampling technique. It was observed that resampling techniques generally increased the success of the classifiers and were more successful when used together with ensemble learning methods. Compared to the other similar studies in the literature, it was shown that the results obtained in this study were higher than the others. [ABSTRACT FROM AUTHOR]

After the COVID-19 epidemic, which emerged in December 2019 and is still in effect, almost all countries had to implement strict measures to control the spread of the virus. The ability of experts to reduce the spread primarily depends on the determination of the criteria affecting the spread. The fact many criteria that affect the rate of spread of COVID-19 and the most effective criteria cannot be determined, causes the spread, and therefore the number of positive cases and deaths to increase. Therefore, in the study; firstly, the weights of the criteria affecting the rate of spread were determined by using the full consistency method (FUCOM), which is a multi-criteria decision-making method, and the criteria that most affected the spread were determined by Pareto analysis, based on the criteria weights obtained. Then, based on the criteria obtained, the number of confirmed cases was predicted using the random forest method. The performance criteria values of the random forest were compared with different artificial intelligence methods such as artificial neural network, decision tree and support vector machine. Random forest gave the best results with error values (RMSE (3247), MAE (1714) and RRSE (0.374)). In addition, the random forest achieved a high prediction success of 92.9%. [ABSTRACT FROM AUTHOR]

Attacks with malware using the Java language have increased rapidly in recent years. With these increases, the damage that malware can cause to people and institutions has led researchers to develop and test different machine learning techniques to improve and strengthen automatic detection systems. In this study, a hybrid system is proposed for the detection of malicious Jar files, which uses binary particle swarm optimization based feature selection and classification with XGBoost algorithm. While minimizing is achieved in the binary particle swarm optimization algorithm, the random forest algorithm is used in the fitness function. With feature selection, it is aimed to increase speed and performance by reducing computational load on classification algorithm. In proposed model, training and tests were carried out by performing 10-fold cross validation. In detection mechanism made with the XGBoost algorithm, the performance of the model established with Accuracy, Precision, F1-Score, Recall metrics has been demonstrated. In order to evaluate performance of the proposed model, tests were made with AdaBoost, Gradient Boosting, Support Vector Machines, Artificial Neural Networks, Naïve Bayes methods and the results were compared. Experimental results showed that the proposed binary particle swarm optimizationbased feature selection and hybrid model classifying with the XGBoost algorithm was more successful in detecting Jar malware than the compared models with an accuracy rate of 98.04%. [ABSTRACT FROM AUTHOR]

The aim of the study is to calculate the mortality risk of patients diagnosed with Covid-19 using Machine Learning algorithms. In the study, demographic and clinical data of patients admitted to the health facility with the diagnosis of Covid-19 in Atlanta, Georgia, which are published as open access on the web, are used. The mortality risk of the patients is calculated using Machine Learning algorithms called Decision Tree, Random Forest and Adaptive Boost based on the data. It is observed that the demographic and clinical findings of the patients are effective on mortality risks and that the Machine Learning-based prediction modeling created in this direction can be applied with high reliability (Acc=83.5). With the findings obtained, highreliability classification models can be created using Machine Learning methods and decision support modules can be created that can guide clinicians and health professionals in patient prioritization in line with the calculation of mortality risks of patients. By creating web-based modules, a scientific basis is established for health authorities, clinicians and hospital managers to make effective and efficient preparations for bed occupancy planning. Unnecessary health expenditures and patients who are likely to have a relatively mild illness can be prevented from receiving unnecessary treatment. [ABSTRACT FROM AUTHOR]

Nowadays it is important to store sensitive data and restrict its usage only to authorized people or institutions. In general, solutions for Data Leakage Prevention (DLP) ignores malicious attacks on documents and algorithms using fingerprinting and regular expressions are used. However, content-based attacks are successful evading those algorithms. In this paper an algorithm robust against malicious content-based attacks is proposed, which is independent of the attack executed. Transposition, sentence structure alteration, modification, obfuscation attacks are taken into consideration within the scope of paper. N-gram, charactergram, k-skip-n-gram and LSA methods are used in the feature extraction step, for having better classification results under attacks. The extracted features are passed to a Vote Classifier consisting of Support Vector Machine, Random Forest and Multi-Layer Perceptron classifiers. Additionally, the effects of instrumenting Spell-Correction in different steps of the algorithm is evaluated, which is effective against modification attacks. [ABSTRACT FROM AUTHOR]

This study is aimed at measuring the perception created by the COVID-19 vaccines on people around the world using Sentiment Analysis - taking into account the fifth generation (5G) cellular networks, which are expected to bring new capabilities that will create opportunities for society. For this purpose, 25642 tweets were taken from Twitter between October and December 2020 and analyzed using Python software. Accordingly, 36.4% of people worldwide who expressed their opinions on Twitter have a positive perception of 5G and the COVID-19 vaccine, also 35.6% of the tweeters do not have a positive or negative view (neutral) has been observed. However, it was observed that 28% of the people expressed negative opinions. The overall sentiment score is 0.15. Also, in this study, Classification and Regression Trees (CART), Naïve Bayes (NB), k-Nearest Neighbour (KNN), and Random Forest (RF) algorithms were applied. According to the findings, the best results were obtained by NB with 0,7852 precision (P) and 0,7445 accuracy (A) values, KNN with 0,8209 recall (R) value, and RF with 0,7866 F-measure (F) value. [ABSTRACT FROM AUTHOR]

In this study, it is aimed to classify the factors affecting calf diseases with Artificial Neural Networks (YSA), Random Forest Algorithm (RO) and Logistic Regression Analysis (LR), to reveal the usability of these methods and to compare their performances. The research material consisted of the farm records of 54 calves kept in Erciyes University, Agricultural Research and Application Center between 2018-2021. In the statistical analysis, the disease history of the calves was the dependent variable; calves gender, breed, birth season, maternal breed, and maternal lactation number were determined as independent variables. Classification performances were compared with sensitivity, specificity, precision, accuracy, f-measure, Youden index, area under the ROC curve (AUC) and Cohen's kappa coefficient. According to the research results, the most successful classifiers in terms of sensitivity, selectivity, precision, accuracy, F-measure, Youden's index and Cohen's kappa; LR (0.828), ANN (0.947), ANN (0.964), ANN (0.833), ANN (0.857), ANN (0.719), ANN (0.663), respectively. In conclusion, it has been reached that be used classification methods correctly classify the factors affecting calf diseases with a somewhat margin of error and YSA is more successful for all performance values except sensitivity. It is thought that these methods will help determine calf diseases with proactive approaches and prevent economic losses in livestock enterprises. [ABSTRACT FROM AUTHOR]

Heart diseases have an important place among the diseases that people face. There can be many reasons for the occurrence of heart diseases and it can affect the person very much. However, some treatments that can prevent people from ending life due to heart disease. At this point, early diagnosis is of great importance. For this reason, the subject of the study was chosen as the performance analysis of the detection of heart diseases using machine learning algorithms. Although the research on this subject is limited, there is some researches with machine learning method. From machine learning methods; support vector machines, random forest algorithms, artificial neural networks, Naive Bayes, and k-NN were compared and analyzed in this study. [ABSTRACT FROM AUTHOR]

In correct classification of disease diagnosis which variables are analyzed and how results are evaluated should be correctly defined in clinical decision making process as well as in statistical approach. It is aimed to determine the algorithm which has the best classification performance by using two different approaches in this study. The data set was obtained from 232 patients aged≥18 who were admitted to endocrinology outpatient clinic of a public hospital between June-September, 2013. Two different approaches were used to classify diagnosis of diabetes. In the first approach, 18 variables which were found statistically significant in multinomial logistic regression method were utilized; in the second approach, all models were built with 21 clinically significant variables which were determined by expert endocrinologist. Diabetes was classified with supervised machine learning methods; including Naïve Bayes, Bayes network, random forest, decision trees, support vector machine, k-nearest neighbors, artificial neural network and multinomial logistic regression. The performance of models was evaluated with accuracy, Kappa statistics, mean absolute error, root mean squared error, relative absolute error, sensitivity, specificity, precision, F-measure, Matthews correlation coefficient, ROC curve and Youden index. 10-folds cross-validation method was applied to test performance of models; runtimes of each algorithm were calculated. Analyses were performed with WEKA 3.8.2 and R Studio 1.1.383. Generally, random forest algorithm had the best performance with accuracy 84.48% and 81.90%, respectively. Clinical significance should be emphasized as well as statistical significance when choosing correct classification model for diagnosis of diabetes. [ABSTRACT FROM AUTHOR]

COVID-19 infection emerged in December 2019 and was defined as a pandemic by the World Health Organization on March 11, 2020. During the pandemic process, measures such as restrictions in social areas and curfews have been taken. It is important to make estimates for the number of cases during the pandemic process. There are estimation methods used in estimating the number of cases, such as time series analysis, divisional model, and machine learning. In this research, 5 machines as Support Vector Machine Algorithm (SVM), Decision Tree Algorithm (DT), Naive Bayes Algorithm (NB), K-Nearest Neighbor Algorithm (KNN), and Random Forest Algorithm (RF) were used to estimate the number of COVID-19 cases. The prediction performance of the algorithms was determined by the accuracy, sensitivity and precision values. When the PCA method is used, the RF algorithm has the highest accuracy and the SMV algorithm has the highest sensitivity and precision. As a result of using this method, the lowest accuracy was obtained in the NB algorithm, and the lowest sensitivity and precision values were obtained in the NB and DT algorithms. When the PCA method was used, the accuracy, sensitivity and precision values in all algorithms were found at the highest level in shopping centers on Opened/Closed dataset. In the Schools Opened/Closed, Restaurants Opened/Closed and Shopping Malls Opened/Closed data set success rates were found respectively 95%, 88%, and 90% in estimating with the RF algorithm. In the Lockdown data, 85% success rate was achieved in the estimation with the DT Algorithm. [ABSTRACT FROM AUTHOR]

In corporations, software issues and software change demands are forwarded to the Information Technology (IT) unit via a demand management system. The priority information in this system has critical importance to the IT unit. However, the priority decision that is left to the individuals who create the demand records may not always be realistic. For instance, a non-critical and low-priority demand may be created with the highest priority, and this may lead to faulty planning and eventually to customer dissatisfaction. In this work, internal customer demands were classified using text mining techniques and their priorities were predicted. The system was trained and tested with the records extracted from the demand management system of a corporation. After cleaning and preprocessing the raw textual demand data, TF-IDF (Term Frequency -- Inverse Document Frequency) weighting scheme was used when creating the document-term matrix. Several classification algorithms were tested on the data set generated, and the highest performance was obtained by Sequential Minimal Optimization algorithm with 54.1% F-Score. In addition, on the dataset made balanced with oversampling technique, the highest performance was achieved by Random Forest algorithm with 74.5% F-Score. [ABSTRACT FROM AUTHOR]