Lung disease impacts millions globally each year, posing significant risks to youth, the elderly (aged 65 and above), and persons with other medical disorders such as hypertension, diabetes, and obesity. Numerous initiatives have been proposed in recent years to successfully identify and categorize lung disorders using medical images, particularly chest X-rays, utilising imaging software. This paper proposes the development of an advanced autonomous lung disease detection system utilising deep learning to predict overall survival (OS) without requiring accurately marked regions by leveraging patient-level life expectancy information. The proposed approach involves training and evaluating a pre-trained model to identify lung disease from input CT scan pictures. The training dataset is subjected to stochastic thinning, and we use a set of small random subsets called fractional training sets, which collectively make up less than 20% of the training data. Resnet50, Alexnet, and VGG-19 are three distinct pre-trained models that are used to measure prediction accuracy and loss function performance. As demonstrated by the experimental data, the suggested framework surpasses state-of-the-art approaches in terms of total detection accuracy and computing complexity.