Features Selection for Classification of SMILES Codes Based on Their Function

Abstract

The classification of the active compound based on their function is important to be done because most of them are unknown their function. The structure of the active compound can be represented by SMILES code that unique, compact and complete. Preprocessing SMILES code is a crucial task before SMILES codes are classified. In this research, preprocessing is extracting SMILES code into several features. The features must represent patterns or information from SMILES codes because the proper features of the SMILES codes will increase the accuracy of classification results. This paper uses features from SMILES codes directed by an expert and based on the previous research. Features will be normalized and are classified by an efficient and good classification method, Extreme Learning Machine (ELM). The experiment results show that first, adding features will increase the average of the accuracy of the system until 10.9% on dataset 1-3-4 (nerve-bacterial-cancer). Second, ELM is superior to SVM and KMNB in terms of both accuracy and processing time.