In this study, we propose a text classification model that addresses the challenge of imbalanced text data using a combination of random oversampling, Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM), and a self-attention mechanism using Transformer. The model aims to achieve high accuracy in classifying imbalanced text data by leveraging various techniques. To handle the class imbalance issue, random oversampling is applied to generate synthetic samples from the minority class. This technique balances the class distribution and ensures that the model learns from both majority and minority classes effectively. The model architecture combines CNN and LSTM layers to capture local and global dependencies in the text. The CNN layer applies filters of different sizes to extract local features and detect patterns at various scales. The LSTM layer captures long-term dependencies and sequential information in the text, enabling the model to understand the context and structure of the input. Furthermore, the model incorporates a self- attention mechanism using the Transformer architecture. This mechanism allows the model to focus on important words or phrases within the text, enhancing its ability to recognize crucial information and make accurate predictions. Experimental results on a benchmark dataset demonstrate the effectiveness of the proposed model. It achieves an impressive accuracy of 98.08%, outperforming previous approaches for imbalanced text classification. The model's high accuracy can be attributed to the combination of random oversampling, CNN-LSTM layers, and the self-attention mechanism, which collectively address the challenges posed by imbalanced text data.