IJCSNS International Journal of Computer Science and Network Security, VOL.7 No.7, July 2007 194 Manuscript received July 5, 2007 Manuscript revised July 25, 2007 Using Support Vector Machines to Enhance the Performance of X-Ray Diffraction Data Analysis in Crystalline Materials Cubic Structure Identification Mohammad Syukur Faculty of Mathematics and Natural Sciences, University of Sumatera Utara, 20155 Medan, Sumut, Indonesia Summary Crystalline materials cubic structure identification is very important in crystallography and material science research. For a long time researchers in the field have used manual approach in matching the result data from X-Ray Diffraction (XRD) method with the known fingerprint. These manual matching processes are complicated and sometimes are tedious because the diffracted data are complex and may have more than one fingerprint inside. This paper proposes the use of support vector machines to enhance the performance of the matching process between the diffracted data of crystalline material and the fingerprints. It is demonstrated, through experiments, that support vector machines gives more accurate and reliable identification results compared to the use of neural network. Key words: Support Vector Machine, X-Ray Diffraction Data Analysis, Cubic Structure Identification, Material Sciences, Artificial Intelligence Application. 1. Background Crystallography is one of the areas of research in physics that deals with the scientific study of crystals. It has always been one of the most challenging research fields since eighteenth century. The significant discovery of X- ray by Röntgen in 1895 [8] had yielded to a new way of doing crystallography research. Since then X-ray diffraction method has been proposed and applied to many different sub-area of crystallography such as identification of crystalline phases, qualitative and quantitative analysis of mixtures and minor constituents, distinction between crystalline and amorphous states, side-chain packing of protein structures, identification of crystalline material, etc. The last area is the focus of this paper. X-ray diffraction data interpretation for most crystalline materials is a very complex and difficult task. This is due to the condition that different crystalline material may contain more than one cubic structures component type and after being diffracted using X-ray diffraction method, the diffracted data are complex. Hence, the data can be very ambiguous and is not easy to track and understand. Numerous artificial intelligence techniques and application have been applied and developed to solve the problems in various domains. In our previous work [5], an attempted has been made to use neural network to perform automatic cubic structure identification on the crystalline materials. Though it was a success, there’s still left room for improvement. This paper proposes the use of support vector machine (SVM) to enhance the performance of crystalline materials cubic structure identification. The result of using SVM is compared with the result using neural network. Support vector machines have been proven to be a powerful method to solve identification and classification problems. It includes gene identification [10], paraphrase identification [11], protein classification using X-ray crystallography [9], and many more. The main intent of this paper is to showcase the superior results on the use of support vector machine over neural network in crystalline materials cubic structure identification. 2. Cubic Structure Identification using X-Ray Diffraction Data In principle, there are four cubic structures type for crystalline materials, the Simple Cubic (SC), Body Centered Cubic (BCC), Face Centered Cubic (FCC) and Diamond [1]. In our previous work [2], a formula has been proposed to calculate the fingerprints for these four cubic structures. The formula utilizes the Miller index (h,k,l) [6]. The proposed formula can be written as follows: 2 2 2 2 2 2 4 ) ( sin a l k h + + = λ θ (1) Since the wavelength of the incoming X-ray (λ) and lattice constant (a) are both constants, we can eliminate these