Detection of adulterants in medicinal products by Infrared spectroscopy and ensemble of window extreme learning machine

Abstract Motivated by the superiority of extreme learning machines (ELM), a so-called simple ensemble version, abbreviated as EWELM (ensemble window ELM), and was devised for multivariate calibration. In such an algorithm, the self-contained random operation of ELM provides an underlying way of generating diversity for designing ensemble algorithms. Fast learning speed of ELM makes the algorithm to be of practical value. A total of 41 samples of Jiangtangning capsules adulterated with metformin hydrochloride are prepared. The adulteration was performed within a concentration range of 0-29%. Partial least squares (PLS) and the full-spectrum ELM (EFELM) are selected as the reference algorithms. The experimental results proved the superiority of the EWELM algorithm to the references, indicating that it is a feasible screening tool to detect synthetic drug adulterants in Chinese patent medicine (TCM) on the market.