The capacity to interact with environments, understand them, and make judgments on time defines smartness, the foundation of smart cities, and civilizations. The main motivation of this study is to satisfy the need for a real-time disaster-related application that increases the demand for novel techniques that are scalable with big data. The main aim of this paper is to analyze the input data to find the crucial features and accurately classify them into their appropriate disaster class with the help of social media. The disaster dataset contains numerous features which increase the dimensionality of the dataset. The existing techniques consume higher runtime memory for large training datasets and suffered from different drawbacks such as oversampling, computational cost, low speed, data imbalance, concept drift, and computational complexity. To overcome these drawbacks, this study presents a novel city councils evolution (CCE)-optimized ensemble support vector machine-based extreme learning machine (ESVM-ELM) model on Apache Spark for predicting disaster events in big data. The traditional serial processing issue is overcome in this paper using an appropriate parallelization technique which improves the speedup of the model and improves the time taken for classification. The ESVM-ELM model performs well with imbalanced datasets and handles the concept drift problem efficiently. The use of the CCE algorithm for optimizing the ESVM-ELM model offers improved accuracy, a better convergence rate, and minimal computational complexity. The efficiency of our model is demonstrated by validation using the disaster tweets dataset and comparison with the four underlying approaches, namely, naïve Bayes, ELM, FCM, and Log-Based Abnormal Task Detection. The cross-validation method is utilized in this paper to generate an ensemble of ELM classifiers for decision-making utilizing an ESVM-ELM algorithm. The proposed model offers improvements in terms of accuracy, precision, recall, and F-measure values when compared to different baseline models. The experimental results demonstrated the efficiency of the ESVM-ELM model in improving the prediction accuracy, speedup, and scale-up for big data classification with reasonable processing time. © 2023 Elsevier B.V., All rights reserved.