Improved Extreme Learning Machine (ELM) based approach(es) for data analysis in Smart City Applications

Abstract

In today’s era of rapid urbanization and advancements in Information and Communication Technology (ICT), Smart Cities have paved the way for efficient management of resources. In these cities, sensors are strategically placed at various locations to collect real-time data, enabling enhanced monitoring, optimization of resources, and improved decision-making for urban management. These sensors, integrated into the urban infrastructure, facilitate continuous data collection that supports smarter city planning, predictive analytics, and the development of responsive, sustainable environments. The data generated by various sensors needs to be analyzed to provide useful information to the users, but it is a challenging task. This can be achieved using various Machine Learning (ML) and Deep Learning (DL) approaches such as support vector machine (SVM), Convolution neural network (CNN) etc. In recent years, Extreme Learning Machine (ELM) has gained importance in the research domain due to its significant features of no backpropagation, no hyperparameter tuning, no human intervention, and simple architecture. In this research work, three ELM-based hybrid approaches have been proposed for accident severity classification, parking space detection, and bike-sharing demand prediction in Smart Cities. In addition to this, a comprehensive survey work on ELM and different variants of ELM is conducted to understand the various applications of ELM. First approach, ELM-based SVM (E-SVM), utilizes feature mapping of ELM and performs classification using SVM, which contributes towards the decision boundary by maximizing the distance between the hyperplanes. The proposed framework for accident severity classification in Smart Cities outperforms other traditional ML algorithms for a majority of the datasets used in the experimental analysis in terms of accuracy, precision, and F1 score. Second approach, CNN-based ELM (CNN-ELM), uses the best properties of CNN and ELM to identify vacant and occupied parking spaces in Smart Cities. The first step involves feature extraction using CNN, and the second step includes classification using ELM. The proposed approach has been evaluated on publicly available PkLot dataset. It has been experimentally proved that CNN-ELM not only reduces the computational time but also improves the classification accuracy as compared to traditional CNN models. Third approach, Grey wolf optimization-based Incremental ELM (GWO-IELM), has been proposed for predicting the count of shared bikes in Smart Cities. The significant features from the dataset are selected using GWO in the first phase, and then the regression is performed using I-ELM in the second phase. The effectiveness of the proposed approach has been validated on a publicly available London bike-sharing dataset. The experimental results verify the superior performance of the proposed approach as compared to conventional ML approaches in terms of coefficient of determination (R2), mean absolute error (MAE), root mean square error (RMSE), and root mean square log error (RMSLE), respectively.