To propose a new AI-based multi-classification model to detect lung nodules and pneumonia with an enhanced image localization framework. The model employs deep learning techniques to learn the deep features from X-ray images to isolate the affected regions in order to boost the detection and classification accuracy. This new system is developed using the ENN classifier VGG19 and an image localization framework YOLO V8 to detect the potential regions affected by lung cancer or pneumonia for robust classification. EHARQ is the error recovery method during the image acquisition phase and ensemble learning for post-processing validation on the isolated regions to check for low confidence scores. A multi-centric dataset is used to evaluate the performance of the proposed model. This research work utilizes X-ray images taken from Kaggle, which contain 5856 JPEG images, which have three labels, health, lung cancer, and pneumonia, under the male and female categories. 70% of the data are used for training, and 30% are used for testing and validation purposes. The performance is measured using MATLAB, where the results are compared with existing baseline approaches such as EfficientNet-B0, Resnet-50, and 3D-DLCNN models. The suggested model enhances the performance rate of detection and classification with proven results. The ENN classifier with YOLO V8 yields 97.3% accuracy, 98.7% precision, 97.3% recall, 98.8% F1 score, 0.93 TPR, and 0.07 FPR under AUC-ROC which outperforms the prevailing EfficientNet-B0, Resnet-50, and 3D-DLCNN models. Robust discrimination of lung cancer, pneumonia, and healthy classes using deep learning and ILF methods, which helps the clinical experts treat the disease at an early stage. The model overcomes the shortcomings of the existing classifiers in terms of image localization and disease classification.