Author_Institution :
Coll. of Inf. Sci. & Eng., Northeastern Univ., Shenyang, China
Abstract :
To study the effectiveness of classification algorithms in cigarette sensory quality evaluation, chemical components such as total sugar, protein, potassium, etc. are taken as condition attributes, and ID3, C4.5, rough set, BP neural network, support vector machine, and k-nearest-neighbor are adopted to predict cigarette sensory quality index, such as luster, aroma, harmony, offensive odor, irritation and aftertaste. The experimental results show that harmony reaches the best classification accuracy with about 95%, and the effectiveness of luster and offensive odor are slightly below the harmony with 85%-90% by SVM and KNN, while aroma has the worst result. In addition, offensive odor and aftertaste are fairly accurate with about 70%. As a whole, SVM and KNN have the better performance in the prediction of cigarette sensory quality than the other classification algorithms.
Keywords :
backpropagation; data mining; neural nets; pattern classification; production engineering computing; quality management; rough set theory; support vector machines; tobacco products; BP neural network; C4.5; ID3; KNN; SVM; aftertaste; aroma; chemical components; cigarette sensory quality evaluation; cigarette sensory quality index prediction; classification algorithms; data mining; harmony; irritation; k-nearest-neighbor; luster; offensive odor; potassium; protein; rough set; support vector machine; total sugar; Chemicals; Classification algorithms; Educational institutions; Indexes; Neural networks; Support vector machines; Training; Classification Algorithms; Data Mining; Experimental Study; Sensory Quality Evaluation;
