Analysis of Forensic Ballistic Specimens for Firearm Identification Using Supervised Naive Bayes and Decision Tree Classification Technique

Abstract

Every crime gun has a story to tell. Recently, the rise of crime cases involving firearms is quite alarming. In ballistics studies, the cartridge cases left by gunshots are important evidences since they act as “fingerprint” to the firearms used. In this study, the secondary data of Legendre Orthogonal Moment (LOM) for the firing pin impression (FPI) images captured from a total of 747 cartridge cases are used. The bullets were shot from five different pistols of the same model namely, type 9 mm Parabellum Vektor SP1, made in South Africa. The method of moment is used in this study since it is more accurate, robust, and less time consuming as compared to the traditional method via comparison microscope. Preliminary analysis using Pearson correlation shows that the features are highly correlated. Therefore, principal component analysis (PCA) was used to analyze the interrelationship among the features and combine them into eight significant component of features while maintaining the information of the original patterns. PCA has reduced the dimensionality of the features. The classification techniques used are naïve Bayes and decision tree analysis on J48 algorithm using the WEKA software due to the simplicity, ability for high dimensionality inputs, and outperformance of these methods than the other more sophisticated classification techniques. Classification results using the naïve Bayes and decision tree techniques show that about 85 and 78.4 % of the images were correctly classified respectively. Hence, the results demonstrate the potential of using numerical features generated by moment of FPI images for firearms identification.