The focus of this research is to predict the decline in pulmonary function in patients with Idiopathic Pulmonary Fibrosis (IPF) using the OSIC Pulmonary Fibrosis Progression dataset. This dataset integrates anonymized clinical data, baseline CT scans, and temporal Forced Vital Capacity (FVC) measurements collected over 1–2 years, addressing the challenge of combining heterogeneous data to model disease progression effectively. The study aims to bridge gaps in existing literature by developing advanced methodologies for accurate IPF prognosis prediction. Initial experiments employed techniques like ElasticNet and Deep Neural Networks (DNNs), providing a foundational understanding but revealing the need for more robust models. To enhance predictive accuracy, hybrid architectures were implemented, including combinations of DNN with NGBoost, Gradient Boosted Decision Trees (GBDT), and LightGBM (LGBM). LSTM was also implemented for the clinical data. Pre-trained convolutional architectures like ResNet-18 and EfficientNet-b0 were leveraged for CT feature extraction, yielding error metrics of − 6.70 ± 0.29 and 183.68 ± 23.52, demonstrating their efficacy in medical image analysis. Advanced spatial-temporal models, such as CNN-LSTM and LSTM-QRNN, were developed to integrate clinical and sequential FVC data, further improving predictive accuracy. The implementation of FibroCoSANet capitalized on convolutional self-attention mechanisms and hybrid feature fusion, outperforming baseline models. Challenges like irregular FVC intervals and effective data integration were addressed iteratively. This research highlights the potential of hybrid and ensemble learning approaches to improve predictions on metrics like the modified Laplace Log-Likelihood. Future work involves optimizing FibrosisNet, refining ensemble methods, and exploring additional spatial-temporal architectures. The findings contribute a robust framework for IPF prognosis, blending machine learning advancements with clinical insights.