Yazar "Efeoglu, Ebru" seçeneğine göre listele

Listeleniyor 1 - 5 / 5
  • Küçük Resim Yok
    Öğe
    Detection of Hazardous Liquids Using Microwave Data and Well-Known Classification Algorithms
    (Pleiades Publishing Inc, 2020) Efeoglu, Ebru; Tuna, Gurkan
    The recent increase in terrorist attacks realized using liquid explosives has made it important to develop quick and reliable methods that can distinguish between nonhazardous liquids and other liquids that can be used in these explosives. Since the stability and sensitivity properties of microwave systems are high, microwave frequency band is preferred to differentiate hazardous liquids from non-hazardous liquids. In this study, a noncontact system based on electromagnetic response measurements of liquids in microwave frequency band is proposed to develop a classification approach that can be used in liquid scanners. Naive Bayes, linear discriminant analysis, qualitative data analysis, support vector machine, sequential minimal optimization, K-nearest neighbors classification algorithms are used to classify liquids and their classification performances are analyzed. The results of the set of classification experiments prove the success of the proposed measurement method. As the results prove, K-nearest neighbors is the most appropriate classification algorithm for hazardous liquid detection. Since it can be easily implemented and its detection process is fast, a classification system based on the proposed approach can be very useful in airports and shopping malls.
  • Küçük Resim Yok
    Öğe
    Determination of salt concentration in water using decision trees and electromagnetic waves
    (Iwa Publishing, 2022) Efeoglu, Ebru; Tuna, Gurkan
    Salt water adversely affects human health and plant growth. In parallel with the increasing interest in non-contact determination of salt concentration in water, a novel approach is proposed in this study. In the proposed approach, S parameter measurements, which show the scattering properties of electromagnetic waves, are used. First, the relationship between salt concentration in water and permittivity values, a distinguishing feature for liquids, is shown. Then, based on the derived correlations from a set of S parameter measurements, it is shown that the salt concentration in water can be predicted. Finally, after exactly determining the relations of permittivity, salt concentration and S parameter, a system that allows non-contact determination of salt concentration is proposed. Since the proposed system makes its prediction using a classifier, decision tree algorithms are employed for this purpose. In order to evaluate the appropriateness and success of the algorithms, a set of classification experiments were held using various water samples with different levels of salt concentration. The results of the classification experiments show that the Hoeffding tree algorithm achieved the best results and is the most suitable decision tree algorithm for determining the salt concentration of liquids. For this reason, the proposed non-contact approach can be used to determine the salt concentration in water reliably and quickly if its hardware and software components can be embedded into a prototype system.
  • Küçük Resim Yok
    Öğe
    Machine Learning for Predictive Maintenance: Support Vector Machines and Different Kernel Functions
    (Pleiades Publishing Inc, 2022) Efeoglu, Ebru; Tuna, Gurkan
    Predictive maintenance relies on machine learning techniques to learn from historical data and also uses live data to analyse failure patterns. Different from conservative maintenance procedures that generally lead to resource wastage, predictive maintenance can offer optimum resource utilisation and allow predict failures before they occur. Machine learning techniques are essential for automated predictive maintenance; therefore, in this paper the use and effectiveness of support vector machines for predictive maintenance is analysed. As the results show, support vector machines achieve the best performance when linear kernel function is used.
  • Küçük Resim Yok
    Öğe
    Use of Hybrid Clustering and Scattering Parameters for Liquid Classification
    (Istanbul Univ-Cerrahpasa, 2022) Efeoglu, Ebru; Tuna, Gurkan
    With the advancement of technology, the use of machine learning techniques has increased. The need for the prevention of terrorist attacks has brought upon the use of machine learning techniques to explosive detection. Flammable liquids such as alcohol are easily available and widely used in various terrorist attacks. In this study, a new microwave measurement system is developed and a hybrid clustering approach is proposed to classify liquids. With the proposed measurement system, the reflection coefficient (S-11 parameter) of liquids in bottles is measured at room temperature and these measurements are used as inputs by the proposed clustering algorithm. The results obtained using the proposed clustering algorithm are compared with the results obtained using a set of well-known clustering algorithms, that is, K-means, hierarchical clustering, farthest first, and fuzzy C-means, in order to make a fair comparison. The results show that the proposed clustering algorithm provides 100% accuracy and is superior to the well-known algorithms used in this study. The results will enable us to manufacture a low-cost liquid scanner for railway stations and shopping malls as well as small airports. The proposed liquid scanner's design was completed, and the manufacturing phase has been started.
  • Küçük Resim Yok
    Öğe
    The Use of Microwave and K* Algorithm in Determination of Alcohol Concentration in Liquids
    (Pleiades Publishing Inc, 2020) Efeoglu, Ebru; Tuna, Gurkan
    Alcohol is widely used in various fields today. Alcohol overdose and its improper use can cause many health problems. Drinking liquids with high alcohol concentration, especially methanol, causes poisoning, and the use of colognes with high alcohol concentration causes various skin conditions and respiratory diseases. Therefore, for public health it is important to determine the type and concentration of alcohol in beverages and liquid products. In this paper, K* algorithm is used to detect and classify liquids containing high concentration of alcohol by measuring the scattering parameter of the liquids in the microwave frequency band. For this purpose, scatter parameter values of aqueous solutions of ethanol, methanol, 1-propanol and 2-propanol at different concentrations have been used as training sets. Commercial liquids with known ethanol concentrations and liquids with other known alcohol concentrations have been tested to prove the accuracy of the proposed approach. As the results show the proposed approach can classify liquids containing alcohol and their alcohol concentrations successfully and with high accuracy.