As social media platforms are growing rapidly, there is an increasing demand for advanced methods to detect and manage crisis events in real time. This paper addresses this need by introducing a novel framework for crisis event classification from social media data. The dual channel, event categorization, and embedding layers are all parts of the proposed architecture. Leveraging the wealth of user-generated content on platforms like Twitter and Facebook, our framework utilizes a dual-channel architecture comprising Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs), augmented with an attention mechanism. Subsequently, the LSTM network and a CNN receive the embedding vectors as inputs. Exploiting the merits of the convolutional layer for spatial attribute extraction and the LSTM layer for deriving text’s sequence and order-related data. The main reason behind the use of CNN and LSTM in both channels. The attention mechanism enables dynamic focus on salient regions of the input sequence. We demonstrate the effectiveness of our framework through empirical evaluation and comparative analysis, showcasing the precision, recall, and F1-Score as 0.64, 0.53, and 0.55, respectively. It shows an improvement of 11 and 2% concerning precision and F1-score with current state-of-the-art methods for precisely identifying and classifying humanitarian crisis events.