This paper introduces an enhanced twin support vector machine with a deep learning approach (EPLF-TSVM-DL) for sentence-level sentiment classification. The method aims to improve the accuracy and efficiency of sentiment analysis by addressing the challenges of feature extraction and classification. The EPLF-TSVM-DL combines deep learning techniques, specifically Convolutional Neural Networks (CNN), with the EPLF-TSVM framework, which uses a truncated pinball loss function to minimize computational complexity and improve generalization. The proposed method is evaluated on two datasets: Sentiment140 and COVID-19_Sentiments, and compared with other existing methods such as SVM, EESNN-SA-OPR, and SWOANN. The experimental results show that the EPLF-TSVM-DL outperforms these methods in terms of accuracy, precision, F1-score, recall, and stability. The study highlights the effectiveness of the proposed approach in handling large-scale sentiment analysis tasks, particularly in low-resource languages like Urdu. Future work includes integrating bidirectional LSTM networks and fuzzy logic theory to further enhance the model's performance.This paper introduces an enhanced twin support vector machine with a deep learning approach (EPLF-TSVM-DL) for sentence-level sentiment classification. The method aims to improve the accuracy and efficiency of sentiment analysis by addressing the challenges of feature extraction and classification. The EPLF-TSVM-DL combines deep learning techniques, specifically Convolutional Neural Networks (CNN), with the EPLF-TSVM framework, which uses a truncated pinball loss function to minimize computational complexity and improve generalization. The proposed method is evaluated on two datasets: Sentiment140 and COVID-19_Sentiments, and compared with other existing methods such as SVM, EESNN-SA-OPR, and SWOANN. The experimental results show that the EPLF-TSVM-DL outperforms these methods in terms of accuracy, precision, F1-score, recall, and stability. The study highlights the effectiveness of the proposed approach in handling large-scale sentiment analysis tasks, particularly in low-resource languages like Urdu. Future work includes integrating bidirectional LSTM networks and fuzzy logic theory to further enhance the model's performance.