<p>The rapid growth of social media data has driven the development of stance detection models, crucial for understanding online opinions. However, their vulnerability to bias, particularly concerning sensitive attributes like gender, is a major challenge. Current research often uses limited features, such as stance, text, and target attributes, potentially leading to the perpetuation of biases. Similarly, gender prediction methods with narrow feature sets highlight the need for more comprehensive approaches to reduce bias. This paper proposes a multi-label deep learning hybrid model for stance detection (MDLSD) that enhances stance detection performance by incorporating gender features. The proposed model predicts both stance and gender using a multi-label learning approach, effectively capturing their interdependencies. The performance of the proposed model was evaluated using a benchmark dataset, and two new datasets were collected with the snscrape tool. To prevent overfitting, the model employs <i>L</i>2 regularization during parameter optimization, minimizing the difference between predicted and actual stance labels. The MDLSD’s performance was evaluated using accuracy and F1-Score and compared against state-of-the-art techniques. It achieved a 2.76% increase in accuracy and a 0.0277 improvement in F1-Score on <i>D</i>2, and a 2.53% accuracy increase and a 0.0315 F1-Score improvement on <i>D</i>3. These experimental results demonstrate that incorporating gender information significantly enhances stance detection performance, establishing the MDLSD model as a robust tool for multi-label classification tasks in textual data.</p>