Lung cancer is one among the most detrimental cancers worldwide, necessitating the enrichment of precise and effective diagnostic techniques critical. The research work presents pioneering techniques for diagnosing lung cancer, with efforts in enhancing the performance of computer-aided diagnostic (CAD) systems to tackle challenges such as incorrect positives (FP) and lung knob separation. Conventional supervised learning approaches frequently struggle in providing acceptable results owing to the obligation of extensive labeled data, which is typically restricted in medical imaging. To overcome this inadequacy, the research here incorporates semi-supervised and unsupervised learning techniques, which in turn exploit both unlabeled and labeled data. Semi-supervised learning learns and improves model accurateness by merging a minor set of labeled examples with a bigger set of unlabeled data, while unsupervised learning realizes essential arrangements and clusters within the data, assisting in the identification of lung malignancy. The proposed structure utilizes the 3D convolutional neural networks (CNN) and multi-task learning algorithms, premeditated to simultaneously work on lung nodule segmentation and FP reduction. Wide-ranging testing on the LUNA16 dataset and Kaggle dataset has shown encouraging results, demonstrating the model’s effectiveness in handling data scarceness and refining diagnostic performance. The work here delivers significant understandings into how semi-supervised and unsupervised learning boost the exactness and competence of lung cancer screening and diagnosis, thus contributing to the use of CAD systems in clinical practice.