Recognition of Explosive Precursors Using Nanowire Sensor Array and Decision Tree Learning

This paper aims to recognize explosive precursors using a nanowire sensor array and decision tree learning algorithm. The nanowire sensor array consisting of tin oxide sensors with four different additives, platinum (Pt), copper (Cu), indium (In), and nickel (Ni) was designed, fabricated and tested using the vapors from four explosive precursors, acetone, nitrobenzene, nitrotoluene, and octane, at eight different concentration levels. A novel pattern recognition technique based on decision tree learning was applied to classify the explosive precursors and estimate their concentration. Classification and regression tree (CART) algorithm was used for classification. The CART was also utilized for the purpose of structure identification in Sugeno fuzzy inference system for estimating the concentration of the precursors. Two CARTs were trained and their testing results were investigated. The decision tree based classifier and concentration estimator showed good recognition rates with an accuracy of 93.75% by CART and an average percent error rate of less than 4%, respectively.