Lung cancer, also known as bronchitic carcinoma is one of the significant cancer type that is causing severe life hazards in both men and women across the globe. It is desired to have an early detection and prevention mechanism of this deadly disease before it welcomes severe consequences. Deep learning models are ubiquitous nowadays in health sector for disease prediction, prevention, diagnosis, drug discovery etc. This study propose an attention based deep CNN-LSTM model to classify patients diagnosed with lung cancer using IQ-OTH/NCCD lung dataset. The dataset contains images which are marked as normal, benign, or malignan type. The performance of the proposed attention-based CNN-LSTM architecture is compared with standalone deep CNN and hybrid deep CNN-LSTM model to infer the impact of attention layer when augmented to CNN-LSTM model. It is found that the performance of CNN-LSTM augmented with attention layer is improved. This is due to the fact that attention layers highlights only relevant and useful information from the feature maps while making prediction. The experimental outcome of this study confirms that the proposed attention-based Deep CNN-LSTM architecture is performing better as compared to the Deep CNN and Deep CNN-LSTM models. The proposed architecture achieved an accuracy score of 99.67%, precision score of 98.76%, recall score of 97.89% and f1 score of 98.32%. In future, the system can be made more reliable & effective by considering multi-modal lung cancer prediction system.