Oktay Yildiz

A Multi-objective Disjoint Set Covers for Reliable Lifetime Maximization of Wireless Sensor Networks

An important challenge facing many large-scale surveillance applications is how to schedule sensors into disjoint subsets to maximize the coverage time span. Due to its NP-hard complexity, the problem of finding the largest number of disjoint set covers (DSC) of sensors has been addressed by many researchers. Majority of these studies employs the Boolean sensing model where a sensor covers a target if it lies within its sensing range. In reality, however, the sensing reliability may be affected by several parameters, e.g., strength of the generated signals, environmental conditions and the sensor’s hardware. To the best of our knowledge, improving coverage reliability of Wireless Sensor Networks (WSNs) has not been explored while solving DSC problem. This paper addresses the problem of improving coverage reliability of WSNs while simultaneously maximizing the number of DSC. Thus, in the context of WSNs design problem, our main contribution is to turn the definition of single-objective DSC problem into a multi-objective problem (MOP) by adopting an additional conflicting objective to be optimized. Specifically, we investigate the performance of two multi-objective evolutionary algorithms in terms of diversity and quality of the Pareto optimal set for the modeled MOP. The simulation results indicate that multi-objective approach results in achieving reliable coverage and large number of DSC compared to a single-objective approach.

Music genre classification with machine learning techniques

The aim of this work is to predict the genres of songs by using machine learning techniques. For this purpose, feature extraction is done by using signal processing techniques, then machine learning algorithms are applied with those features to do a multiclass classification for music genres.

Forecasting the annual electricity consumption of Turkey using a hybrid model

In this study, we implemented traditional, artificial intelligence and hybrid methods to predict electricity consumption of Turkey. While traditional method is multiple linear regression and artificial intelligence method is artificial neural network, hybrid method is a new method combining these two methods. The data used in the study was provided from Turkish Electricity Transmission Company, Turkish Electricity Distribution Company, Ministry of Energy and Natural Resources and Turkish Statistical Institute which are the public institutions in Turkey. The performance was evaluated using mean absolute percentage error (MAPE), mean square error (MSE) and root mean square error (RMSE). The test of the proposed hybrid model resulted in an average absolute forecast error of 2.25 percent.

Performance analysis of machine learning techniques in intrusion detection

With computer and Internet to be an indispensable part of our daily lives, the number of Web applications on the Internet has increased rapidly. With the increasing number of Web applications, attacks on the disclosure of data on the internet and the number of varieties has increased. Made over the Web attacks and to detect unauthorized access requests, intrusion detection systems have been used successfully. In this study, In order to develop a more efficient STS, machine learning techniques, Bayesian networks, support vector machines, neural networks, k nearest neighbor algorithm and decision trees examined the success of the STS, the success and process time of the classifier according to the types of attacks have been analyzed. Kddcup99 data sets were used in experimental studies.

Development of content based book recommendation system using genetic algorithm

Trend of internet, makes presentation of right content to right user will inevitably. For this purpose, recommendation systems are widely used for the areas of music, book, movie, touristic travel plannig, e-commerce, education and many more. The approach of recommender systems are based on the ground of users history of choices, likings and reviews, each of which is interpreted in order to project the future choices of the user. In this study, a novel and strong recommender system for the books is proposed. A content based book recommendation application was developed which makes recommendations according to users taste and choices.

Cardiac arrhythmia analysis using Hidden Markov Model and murmur diagnosis

The heart is the most important of the vital organs and heart diseases may cause fatal consequences. Abnormal heart sounds called murmur may be a precursor of many serious heart diseases. Cardiac auscultation is a basic technique to easily diagnose murmur disease. Auscultation can be supported by using computer aided automatic diagnosis systems to fast and accurate diagnosis. These systems are useful to remote diagnosis systems in place that have the lack of physicians and modern techniques. In this study, heart sound data taken from different patients and recorded with the help of an electronic stethoscope are classified by using Hidden Markov Model. At the end of this study, healthy heart and five different murmur diseases; can be detected with full success just listening heart sound by an automatic system.

Feature selection and dimensionality reduction on gene expressions

Breast cancer is the most common type of cancer among women. Early diagnosis of the breast cancer plays an important role in treating the disease. Thousands of genes microarray data is often used in cancer diagnosis. However, many of these genes which are used in the diagnosis of disease do not have a meaningful pattern. Also, to classify thousands of genes are not good in terms of performance. Therefore, it is very important to make a correct diagnosis with a small number of genes. In this study, Fisher correlation score and T test were firstly applied for gene selection. After filtering, three different approaches were applied. The first method is feature generation and dimensionality reduction with principal component analysis. The second method is feature generation and feature selection with discrete cosine transform. The third method is feature selection with filtering data.

Traffic planning application made by using artificial intelligence (TPAUAI)

Every year, more than ten thousand people die because of traffic accidents in Turkey and about two hundred thousand people are injured. After examining the traffic accident statics, 90% of the accident are caused by drivers mistakes. Drivers who are playing important role in the occurence of traffic accidents, must be aware of situtations of roads and risky areas on their routes during journey, not only ensures the safe journeys but also helps in running traffic safely. There are lots of products which calculate the route, determine the locations etc. those products have been developing day by day to provide more qualified services.In this project, by using data of traffic accident fact-finding report belonging to Ankara province, determination of risky areas implementation artificial intelligence-based is developed. In this application artificial neural Networks is being used; displaying risk areas on the map depend on weather situtation, time, vehicle type parameters and showing alternative way has been provided.

Gene selection for breast cancer

Breast cancer can be fatal and so it is very dangerous. Early diagnosis of breast cancer has been playing very important role on treatment of the disease. Recently, gene technology has been widely used in cancer diagnosis. A microarray is a tool for analyzing gene expression. Microarray data usually contain thousands of genes and a small number of samples. Although, most of them are irrelevant or insignificant to a clinical diagnosis. It is very difficult to obtain a satisfactory classification result by machine learning techniques because of both the curse-of dimensionality problem and the overfitting problem. Therefore, feature selection plays a crucial role in microarray analysis. In this paper, significant biomarker genes for diagnosis have been identified by feature selection. We attempted to use these markers for the classification of breast cancer. Subsequently, SVM was also used to verify the classification rate of genes selected by feature selection. The classification rate of SVM reaches to 82.69% when using selected genes.

Airborne Multi Function Electronically Scanned Array Radar Simulation

Multifunction (multimode) phased array radars are used for ground, air or maritime surveillance. Electronic scanning feature of these radars provide fast and accurate beam direction as well as short sampling intervals. In this paper, simulation of an airborne multifunction electronically scanned array (ESA) radar (MFESA) is presented