Denoising Autoencoder - Extreme Learning Machine for Improving the Classification Performance Integrated with IoT Water Quality Systems
Muhammad Thoriq Arkaan Susila, Ilham Nadiyansyah Firdaus, Muhammad Farid Chuzairi, Ig. Prasetya Dwi Wibawa, Meta Kallista · 2023
Extreme learning machines (ELM) are fast, accurate, and require less user interference. However, this method cannot handle missing values in the data. By using the autoencoder (AE), missing value imputation can be handled. Combining these two methods provides a better evaluation of accuracy using some open-source datasets. This paper focuses on combining machine learning (ML) methods, just like combining denoising autoencoders and extreme learning machines (DAE-ELM), to provide better results. The proposed method is implemented in real-world problems, i.e., water quality monitoring in Tirtawening, which is one of the Indonesian regional water utility companies located in Bandung, West Java, Indonesia. Using DAE for missing value imputation and ELM classification can increase the performance improvement rate (PIR%) by 2.01%. The DAE-ELM combination improves the accuracy of drinking water quality datasets, with the accuracy increasing by 2.71% from 95.98% to 98.69%.