Abstract: Lung cancer, as one of the malignant tumors with the highest incidence and mortality rates worldwide, faces three core challenges in clinical diagnosis and treatment: insufficient sensitivity in early detection, inadequate individualization of treatment plans, and limited precision in prognosis assessment. These obstacles severely impede improvements in patient survival outcomes. Machine learning (ML), with its robust capabilities in data mining and pattern recognition, offers an innovative pathway to integrate multidimensional medical data and overcome traditional diagnostic and therapeutic bottlenecks. This review adopts a data-model-application framework to systematically examine the dimensional characteristics and clinical value of multi-source data. It delves into the applicability and technical features of traditional ML, deep learning (DL), multimodal fusion (MF), and emerging models, systematically outlining their research advancements in early lung cancer diagnosis, treatment optimization, and prognosis assessment. By summarizing the collaborative mechanisms and model selection strategies across different data sources, it reveals the intrinsic logic that data characteristics determine model selection, and model performance underpins clinical value. Finally, addressing key bottlenecks such as insufficient data standardization, limitations in multimodal fusion technology, lack of algorithmic interpretability, and delayed clinical translation, corresponding solutions are proposed. These provide theoretical support and technical references for advancing the deep integration of ML with lung cancer clinical decision-making and achieving precision medicine transformation.