Alper Kursat Uysal

A novel probabilistic feature selection method for text classification

High dimensionality of the feature space is one of the most important concerns in text classification problems due to processing time and accuracy considerations. Selection of distinctive features is therefore essential for text classification. This study proposes a novel filter based probabilistic feature selection method, namely distinguishing feature selector (DFS), for text classification. The proposed method is compared with well-known filter approaches including chi square, information gain, Gini index and deviation from Poisson distribution. The comparison is carried out for different datasets, classification algorithms, and success measures. Experimental results explicitly indicate that DFS offers a competitive performance with respect to the abovementioned approaches in terms of classification accuracy, dimension reduction rate and processing time.

The impact of preprocessing on text classification

Preprocessing is one of the key components in a typical text classification framework. This paper aims to extensively examine the impact of preprocessing on text classification in terms of various aspects such as classification accuracy, text domain, text language, and dimension reduction. For this purpose, all possible combinations of widely used preprocessing tasks are comparatively evaluated on two different domains, namely e-mail and news, and in two different languages, namely Turkish and English. In this way, contribution of the preprocessing tasks to classification success at various feature dimensions, possible interactions among these tasks, and also dependency of these tasks to the respective languages and domains are comprehensively assessed. Experimental analysis on benchmark datasets reveals that choosing appropriate combinations of preprocessing tasks, rather than enabling or disabling them all, may provide significant improvement on classification accuracy depending on the domain and language studied on.

An improved global feature selection scheme for text classification

An improved global feature selection scheme is proposed for text classification.It is an ensemble method combining the power of two filter-based methods.The new method combines a global and a one-sided local feature selection method.By incorporating these methods, the feature set represents classes almost equally.This method outperforms the individual performances of feature selection methods. Feature selection is known as a good solution to the high dimensionality of the feature space and mostly preferred feature selection methods for text classification are filter-based ones. In a common filter-based feature selection scheme, unique scores are assigned to features depending on their discriminative power and these features are sorted in descending order according to the scores. Then, the last step is to add top-N features to the feature set where N is generally an empirically determined number. In this paper, an improved global feature selection scheme (IGFSS) where the last step in a common feature selection scheme is modified in order to obtain a more representative feature set is proposed. Although feature set constructed by a common feature selection scheme successfully represents some of the classes, a number of classes may not be even represented. Consequently, IGFSS aims to improve the classification performance of global feature selection methods by creating a feature set representing all classes almost equally. For this purpose, a local feature selection method is used in IGFSS to label features according to their discriminative power on classes and these labels are used while producing the feature sets. Experimental results on well-known benchmark datasets with various classifiers indicate that IGFSS improves the performance of classification in terms of two widely-known metrics namely Micro-F1 and Macro-F1.

Text classification using genetic algorithm oriented latent semantic features

Abstract In this paper, genetic algorithm oriented latent semantic features (GALSF) are proposed to obtain better representation of documents in text classification. The proposed approach consists of feature selection and feature transformation stages. The first stage is carried out using the state-of-the-art filter-based methods. The second stage employs latent semantic indexing (LSI) empowered by genetic algorithm such that a better projection is attained using appropriate singular vectors, which are not limited to the ones corresponding to the largest singular values, unlike standard LSI approach. In this way, the singular vectors with small singular values may also be used for projection whereas the vectors with large singular values may be eliminated as well to obtain better discrimination. Experimental results demonstrate that GALSF outperforms both LSI and filter-based feature selection methods on benchmark datasets for various feature dimensions.

A novel framework for SMS spam filtering

A novel framework for SMS spam filtering is introduced in this paper to prevent mobile phone users from unsolicited SMS messages. The framework makes use of two distinct feature selection approaches based on information gain and chi-square metrics to find out discriminative features representing SMS messages. The discriminative feature subsets are then employed in two different Bayesian-based classifiers, so that SMS messages are categorized as either spam or legitimate. Moreover, the paper introduces a real-time mobile application for Android™ based mobile phones utilizing the proposed spam filtering scheme, as well. Hence, SMS spam messages are silently filtered out without disturbing phone users. Effectiveness of the filtering framework is evaluated on a large SMS message collection including legitimate and spam messages. Following the evaluation, remarkably accurate classification results are obtained for both spam and legitimate messages.

ECG based biometric authentication using ensemble of features

In this work, the efficacy of various features on electrocardiogram (ECG) based biometric authentication process is thoroughly examined. In particular, the features acquired from temporal analysis, wavelet transformation, power spectral density estimation and QRS-complex detection over ECG signals are considered. These features are employed with two distinct classification algorithms, namely decision tree and Bayes network, specifically for gender, age and identity recognition problems. The biometric authentication framework is evaluated on a benchmark dataset that contains ECG records of 18 healthy people including 5 men, aged 26 to 45, and 13 women, aged 20 to 50. The results of the experimental analysis reveal that if all those features are used in combination rather than individually, better performance is attained for all classifiers in each recognition problem.

Detection of SMS spam messages on mobile phones

In this study, a novel “SMS spam message filter” utilizing effective feature selection and pattern classification techniques is proposed. The proposed filter detects and filters out SMS spam messages in a smart manner rather than black/white list approaches that require intervention of phone users. In the study, Gini index based approach is preferred as the feature selection method. The feature vectors consisting of the selected discriminative features are then fed into two well-known pattern classifiers, namely Naive Bayes and k-Nearest Neighbor, for recognition process. Furthermore, a mobile application, which exploits the proposed detection scheme, is developed particularly for the mobile phones with Android™ operating system. Thus, SMS spam messages are automatically filtered out without disturbing the phone user. The proposed detection scheme is evaluated on a large SMS message dataset consisting of spam and legitimate messages. The results of the experimental work reveal that the proposed system is considerably successful in filtering SMS spam messages.

Classification of medical documents according to diseases

Medical text classification is still one of the popular research problems inside text classification domain. Apart from some text data compiled from hospital records, most of the researchers in this field evaluate their classification methodologies on documents from MEDLINE database. When whole documents in the database are taken into consideration, MEDLINE is a multi-class and multi-label database. A dataset, containing a small subset of MEDLINE documents belonging to disease categories, is constructed in this study. It is a multi-class but single-label dataset. Due to the highly unbalanced distribution of this dataset, only documents belonging to top-10 disease categories are used in the experiments. The performances of three different pattern classifiers are analyzed on disease classification problem using this dataset. These three pattern classifiers are Bayesian network, C4.5 decision tree, and Random Forest trees. Experiments are realized for the two different cases where the stemming preprocessing step is applied or not. Experimental results show that the most successful classifier among three classifiers is Bayesian network classifier. Also, the best performance is obtained without applying stemming.

ECG classification using ensemble of features

In the literature, countless efforts have been made to analyze and classify electrocardiogram (ECG) signals belonging to various heart problems. In all these efforts, many feature extraction strategies have been used to expose discriminative information from ECG signals. In this paper, the contributions of widely used features to the classification performance and the required processing times to extract those features are comparatively analyzed. The utilized features can be briefly listed as time domain (TD), wavelet transform (WT), and power spectral density (PSD) based features. These feature sets are employed individually and in combination within well-known pattern classifiers, namely decision tree and artificial neural network, to assess classification performance in each case. Later, a wrapper-based feature selection strategy is used to reveal the most discriminative feature subset among the entire feature set containing all the three previously mentioned feature sets. The proposed framework is assessed considering four classes of heart conditions including normal, congestive heart failure, ventricular tachyarrhythmia and atrial fibrillation. The results of the experiments conducted on a large dataset reveal that appropriate subset of TD, WT, and PSD features rather than individual features offer higher classification performance. On the other hand, if the processing time is of concern, TD features come out on top with moderate classification performance.

A survey on ECG analysis

Abstract The electrocardiogram (ECG) signal basically corresponds to the electrical activity of the heart. In the literature, the ECG signal has been analyzed and utilized for various purposes, such as measuring the heart rate, examining the rhythm of heartbeats, diagnosing heart abnormalities, emotion recognition and biometric identification. ECG analysis (depending on the type of the analysis) can contain several steps, such as preprocessing, feature extraction, feature selection, feature transformation and classification. Performing each step is crucial for the sake of the related analysis. In addition, the employed success measures and appropriate constitution of the ECG signal database play important roles in the analysis as well. In this work, the literature on ECG analysis, mostly from the last decade, is comprehensively reviewed based on all of the major aspects mentioned above. Each step in ECG analysis is briefly described, and the related studies are provided.