18 research outputs found
Data mining through data visualization: a case study on predicting churners on telecomunications data set
No full textBaşarslan, Muhammet Sinan (Dogus Author)Data mining is the process of extracting meaningful information from a large, raw data. These processes are carried out by various, detailed methods. And, the obtained results are used to make various interpretations and to draw conclusions. Deductions can either be made by interpreting the data after various operations or by plotting the data in various forms of graphs. This type of interpretation over graphics is called data mining through data visualization. Generating graphs that can be used to draw various conclusions on a telecommunications data set with the help of some packages included in the R program is presented in the paper. It does not require upper-level math skills to interpret these graphics; and everyone having knowledge about the industry and data set of the graphs has the ability to plot similar graphs and make analysis and interpretations regarding the results obtained on the data set at hand. In this study, R language was preferred as the software infrastructure for data mining applications, and graphs were plotted for interpretation through data visualization with data mining
Performance Analysis Of Fuzzy Rough Set-Based And Correlation-Based Attribute Selection Methods On Detection Of Chronic Kidney Disease With Various Classifiers
No full textInternational Scientific Meeting on Electrical-Electronics and Biomedical Engineering and Computer Science (EBBT) -- APR 24-26, 2019 -- Istanbul Arel Univ, Kemal Gozukara Campus, Istanbul, TURKEYBASARSLAN, MUHAMMET SINAN/0000-0002-7996-9169WOS: 000491430200016Technological developments generally have positive effects on our daily lives especially on health domain. Diagnosing diseases through new machines or methods are easier than compared to the past. Benchmarking the effect of attribute selection methods on the performance of classification algorithms in a study to diganose the chronic kidney disease (CKD) by using classification algorithms are aimed. Data set on CKD taken from the UCI machine learning repository has been used for the experiments. After a variety of pre-processing, normalization and attribute selection processes, classifier models are designed. In order to determine the attributes that have gerater contribution on the classification results, the Correlation Based attribute selection (CBAS) method and Fuzzy Rough Set Based attribute selection (FRSBAS) method were used. Two data sets obtained by each attribute selection method and the raw data are classified by 4 classifiers including k-Nearest Neighbor, Navie Bayes, Random Forest and Logistic Regression. The test and training data are separated by 5-fold cross validation. The accuracy, precision, sensitivity, ROC curve and F-measure parameters obtained from confusion matrix are used to compare and evaluate the results of the models. As a result of the study, it is seen that the application of FRSBAS method on CKD data set performs better in all classification algorithms.IEEE Turkey Sect, IEEE EMB, Erasmus+, Europas
Bir banka müşteri verilerinin farklı veri madenciliǧi platformlarında sınıflandırılması
No full textBaşarslan, Muhammet Sinan (Dogus Author) -- Conference full title: 4th Electric Electronics, Computer Science, Biomedical Engineerings' Meeting, EBBT 2018; Tepekent Campus in Istanbul Arel UniversityIstanbul; Turkey; 18 April 2018 through 19 April 2018.The process of extracting meaningful rules from big and complex data is called data mining. Data mining has an increasing popularity in every field today. Data units are established in customer-oriented industries such as marketing, finance and telecommunication to work on the customer churn and acquisition, in particular. Among the data mining methods, classification algorithms are used in studies conducted for customer acquisition to predict the potential customers of the company in question in the related industry. In this study, bank marketing data set in UCI Machine Learning Data Set was used by creating models with the same classification algorithms in different data mining programs. Accuracy, precision and f- measure criteria were used to test performances of the classification models. When creating the classification models, the test and training data sets were randomly divided by the holdout method to evaluate the performance of the data set. The data set was divided into training and test data sets with the 60-40%, 75-25% and 80-20% separation ratios. Data mining programs used for these processes are the R, Knime, RapidMiner and WEKA. And, classification algorithms commonly used in these platforms are the k-nearest neighbor (k-nn), Naive Bayes, and C4.5 decision tree
Açık kaynak kodlu veri madenciliği programları: R’da örnek uygulama
No full textBaşarslan, Muhammet Sinan (Dogus Author)The processes on the way from raw data to meaningful information is called data mining. The data is processed by applying various methods of data mining in order to extract hidden information among raw data. The processed raw data becomes usable in the next steps of data mining. There are many open source and commercial applications to be used in data mining and data processing. In this study, information about data mining programs are given, and a case study on the R program. The R program has been chosen because it has a large preference rate among the users as shown by various graphs.Ham verilerden anlamlı bilgilere geçiş sürecine veri madenciliği denir. Veri, ham veriler arasında gizli bilgileri çıkarmak için çeşitli veri madenciliği yöntemleri uygulanarak işlenir. İşlenmiş ham veriler, veri madenciliğinin bir sonraki aşamasında kullanılabilir hale gelir. Veri madenciliği ve veri işlemede kullanılmak üzere birçok açık kaynak ve ticari uygulama vardır. Bu çalışmada veri madenciliği programları hakkında bilgi verilmiş ve R programı üzerinde bir vaka çalışması sunulmuştur. R programı, çeşitli grafiklerle de gösterildiği üzere kullanıcılar arasında büyük bir tercih oranına sahip olması dolayısıyla seçilmiştir
TSCBAS: a novel correlation based attribute selection method and application on telecommunications churn analysis
No full textBaşarslan, Muhammet Sinan (Dogus Author) -- Conference full title: 2018 International Conference on Artificial Intelligence and Data Processing (IDAP); IEEE; Malatya; Turkey; 28 September 2018 through 30 September 2018.Attribute selection has a significant effect on the performance of the machine learning studies by selecting the attributes having significant effect on result, reducing the number of attributes, and reducing the calculation cost. In this study, a new attribute selection method which is a combination of the Rcorrelation coefficient-based attribute selection (RCBAS) and the ρ-correlation coefficient-based attribute selection (ρCBAS) called the Two-Stage Correlation-Based Attribute Selection (TSCBAS) is proposed to select significant attributes. The proposed attribute selection method has been applied to customer churn prediction on a telecommunications dataset for performance evaluation. The dataset used in the study includes real customer call records details for the years 2013 and 2014 obtained from a major telecommunications company in Turkey. Apart from the proposed attribute selection method, four different methods named Rcorrelation coefficient-based attribute selection, ρ-correlation coefficient-based attribute selection, ReliefF, and Gain Ratio have been used for creating five datasets. After that, four classifier algorithms including Random Forest, C4.5 Decision Tree, Naive Bayes and AdaBoost.M1 have been applied. The obtained results have been compared according to the performance metrics comprising Accuracy (ACC), Sensitivity (TPR), Specificity (SPC), F-measure (F), AUC (area under the ROC curve), and run-time. The results of the comparisons show that the proposed attribute selection algorithm outperforms the state of the art methods on customer churn prediction
Sentiment analysis using a deep ensemble learning model
No full textThe coronavirus pandemic has kept people away from social life and this has led to an increase in the use of social media over the past two years. Thanks to social media, people can now instantly share their thoughts on various topics such as their favourite movies, restaurants, hotels, etc. This has created a huge amount of data and many researchers from different sciences have focused on analysing this data. Natural Language Processing (NLP) is one of these areas of computer science that uses artificial technologies. Sentiment analysis is also one of the tasks of NLP, which is based on extracting emotions from huge post data. In this study, sentiment analysis was performed on two datasets of tweets about coronavirus and TripAdvisor hotel reviews. A frequency-based word representation method (Term Frequency-Inverse Document Frequency (TF-IDF)) and a prediction-based Word2Vec word embedding method were used to vectorise the datasets. Sentiment analysis models were then built using single machine learning methods (Decision Trees-DT, K-Nearest Neighbour-KNN, Naive Bayes-NB and Support Vector Machine-SVM), single deep learning methods (Long Short Term Memory-LSTM, Recurrent Neural Network-RNN) and heterogeneous ensemble learning methods (Stacking and Majority Voting) based on these single machine learning and deep learning methods. Accuracy was used as a performance measure. The heterogeneous model with stacking (LSTM-RNN) has outperformed the other models with accuracy values of 0.864 on the coronavirus dataset and 0.898 on the Trip Advisor dataset and they have been evaluated as promising results when compared to the literature. It has been observed that the use of single methods as an ensemble gives better results, which is consistent with the literature, which is a step forward in the detection of sentiments through posts. Investigating the performance of heterogeneous ensemble learning models based on different algorithms in sentiment analysis tasks is planned as future work
A hybrid classification example in the diagnosis of skin disease with cryotherapy and immunotherapy treatmec
No full textBaşarslan, Muhammet Sinan (Dogus Author) -- Conference full title: 2nd International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT); IEEE; Ankara; Turkey; 19 October 2018 through 21 October 2018.Uncontrolled tumors in the human body are called cancer. Unbalanced diet, alcohol and cigarette use, food additives and a variety of viruses can cause people have cancer. Cancercausing tumors can be good or malignant. This study will measure the responses to treatments for skin disease caused by human papilloma virus (HPV), also called wart virus, which is directly related to cancer. This virus is an infectious virus that can infect another person by contact. There are multiple types of HPV virus and although it is usually benign, it can cause cancers such as cervical cancer, skin cancer. Apart from cancer, warts caused by HPV virus are generally seen on hands, feet, face and genital areas. As the skin grows and sagging progresses, it causes cancer at advanced levels. As a treatment method; drug use, surgical removal and HPV virus vaccination are used. These methods may require various surgical interventions. It can also cause a variety of reactions to allergic patients or it can cause a slight dependence on drug use. In addition to these methods, cryotherapy (ice treatment) and immunotherapy methods are used which are developed to obtain faster results and less costly than drugs and surgical interventions. In this study, it was estimated that 180 patients with warts on hands and feet who applied to the dermatology clinic of Ghaem Hospital in Iran were divided into two groups and responded to the treatment with two separate data sets obtained by applying cryotherapy in the other half and immunotherapy treatment in the other half. These data sets are located in the UCI data set. Navie Bayes, C4.5 decision tree, logistic regression, k- nearest neighbor classifier models have been developed for estimation work. In addition, the classification of the features included in the immunotherapy and cryotherapy data sets were tested by applying the feature selection process. The performance of the data sets after attribute selection and the performance of the raw data sets in the classification models are compared. 5 and 10 times cross validation is used to compare the performance of these models. The study also gave the best performance in all the performance criteria of the 4 different classifiers in the two datasets that are used as common models with the C4.5 Decision Tree. In addition, it is clearly seen that the attribute selection process has increased the performance criteria of all models.İnsan vücudunda kontrol dışı gelişen tümörlere kanser denir. Dengesiz beslenme, alkol ve sigara kullanımı, yemeklerdeki katkı maddeleri ve çeşitli virüsler aracılığı ile insanlar kanser hastası olabilmektedirler. Kansere sebep olan tümörler iyi veya kötü huylu olabilir. Bu çalışma da kanser ile direk ilişkisi olan Human papilloma virüsü (hpv) diğer adıyla siğil virüsünün sebep olduğu cilt hastalığına yönelik uygulanan tedavilere verilen tepkiler ölçülecektir. Bu virüs bulaşıcı bir virüs olup insanların ciltlerinin temas etmesi sonucu bir insandan diğerine bulaşır. HPV virüsünün birden çok türü vardır ve genelde iyi huylu olmasına rağmen rahim ağzı kanseri, cilt kanseri gibi kanserlere sebep olabilmektedir. Kanserin dışında HPV virüsünün sebep olduğu siğiller el ve ayaklarda, yüz ve genital bölgelerde görülür. Cilt büyüme ve sarkmasının ilerlemesi ile beraber ileri derecelerde kansere sebep olur. Tedavi yöntemi olarak; ilaç kullanımı, ameliyat ile alınma ve HPV virüs aşısı gibi yöntemler kullanılır. Bu yöntemler çeşitli cerrahi müdahaleler gerektirebilir. Ayrıca ilaç kullanımı ile ilaca bağımlılığa sebep olmasıyla beraber alerjik hastalarda ilaca farklı tepkiler de verilebilir. Bu yöntemlerin dışında daha hızlı ilaç ve cerrahi müdahaleye göre daha az masraflı ve hızlı sonuç almak için geliştirilen kroyoterapi (Buz tedavisi) ve immunoterapi yöntemleri kullanılır. Bu çalışmada İran'daki Ghaem Hastanesinin dermatoloji kliniğine başvuran el ve ayaklarında siğil olan 180 hasta ikiye ayrılarak yarısında kroyoterapi, diğer yarısında immunoterapi tedavisi uygulanarak elde edilen iki ayrı veri seti ile tedaviye cevap vermeleri tahmin edilmiştir. Bu veri setleri UCI veri setinde yer almaktadır. Tahmin çalışması için Navie Bayes, C4.5 karar ağacı, logistik regresyon, k en yakın komşu algoritmaları ile sınıflayıcı modeller oluşturulmuştur. Ayrıca immunoterapi ve kroyoterapi veri setlerinde yer alan özniteliklerin sınıflandırma da etkisi öznitelik seçme işlemi uygulanarak test edilmiştir. Öznitelik seçme işlemi sonrası oluşan veri setleri ile ham veri setlerinin sınıflandırma modellerindeki başarısı karşılaştırılmıştır. Bu modellerin performanslarının karşılaştırılması için 5 ve 10 kat çapraz geçerleme yöntemi kullanılmıştır. Çalışma da ortak olarak kullanılan iki veri setinde de 4 ayrı sınıflayıcı modellerinin tüm performans kriterlerinde diğer modellere göre en iyi performansı C4.5 Karar ağacı ile oluşturulan modeller vermiştir. Ayrıca öznitelik seçme işlemi uygulanan veri setlerinin tüm performans kriterlerinde artışı görülmüştür
Sentiment Analysis with Machine Learning Methods on Social Media
No full textSocial media has become an important part of our everyday life due to the widespread use of the Internet. Of the social media services, Twitter is among the most used ones around the world. People share their opinions by writing tweets about numerous subjects, such as politics, sports, economy, etc. Millions of tweets per day create a huge dataset, which drew attention of the data scientists to focus on these data for sentiment analysis. The sentiment analysis focuses to identify the social media posts of users about a specific topic and categorize them as positive, negative or neutral. Thus, the study aims to investigate the effect of types of text representation on the performance of sentiment analysis. In this study, two datasets were used in the experiments. The first one is the user reviews about movies from the IMDB, which has been labeled by Kotzias, and the second one is the Twitter tweets, including the tweets of users about health topic in English in 2019, collected using the Twitter API. The Python programming language was used in the study both for implementing the classification models using the Naïve Bayes (NB), Support Vector Machines (SVM) and Artificial Neural Networks (ANN) algorithms, and for categorizing the sentiments as positive, negative and neutral. The feature extraction from the dataset was performed using Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec (W2V) modeling techniques. The success percentages of the classification algorithms were compared at the end. According to the experimental results, Artificial Neural Network had the best accuracy performance in both datasets compared to the others
Comparative Analysis of Kolmogorov-Inspired CNN and Traditional CNN Models for Pneumonia Detection: A Study on Chest CT Images
No full textAim: In this study, our goal is to compare the effectiveness of Kolmogorov Inspired Convolutional Neural Networks (KAN) with traditional Convolutional Neural Networks (CNN) models in pneumonia detection and to contribute to the development of more efficient and accurate diagnostic tools in the field of medical imaging.
Methods: Both models are structured with the same layers and hyperparameters to ensure a fair comparison of their performance. For a robust evaluation, the relevant dataset was divided into 80% for training and 20% for testing.
Results and Conclusion: Performance metrics of KAN; 95.2% sensitivity, 97.6% specificity, 94.1% precision, 96.9% accuracy (Acc), 0.9466 F1 score (F1) and 0. 9251 Matthews Correlation Coefficient (MCC), while the CNN model was found 92.5%, 96.4%, 91.2%, 95.3%, 0.9188 and 0.8858 for the same criteria, indicating that KAN outperformed. This comparison emphasizes that KAN has the potential to be a more effective model for pneumonia detection in chest CT images
