This study explores major tumor features to improve predictive modelling for breast cancer diagnosis. We use a dataset that consists of the mean, standard error, and worst values for various attributes including radius, texture, and concavity, and then we analyze the correlation between these features and the corresponding diagnosis outcome (malignant or benign). The decision tree classifier shed light on the critical features contributing to malignant or benign tumors, emphasizing the significance of concave points (worst), perimeter (worst), and area (worst) through feature importance analysis. The key findings indicated that there are very different distributions of the features between malignant and benign cases, especially in radius_mean, area_mean, and smoothness_mean, making them informative of the diagnosis. We can see from the correlation matrix that there are strong relationships between radius_mean, perimeter_mean and area_mean, this shows that we might need to look into dimensionality reductions and eliminate redundancy in the dataset. Diagnostic differences are visualized through pair plots and distribution plots to gain additional insights into improving the network classification accuracy. The significance of targeted feature engineering is evident in this research work, which also lays a foundation for building the best machine learning models for early and accurate detection of breast cancer. The results serve the larger purpose of reducing diagnostic errors and improving patient outcomes.