Shared spectrum environments, essential for future networks, face challenges such as interference and dynamic usage. The paper explores the application of machine learning (ML) techniques to the NIST RF dataset, a synthetic collection of radar signals in the 3.5 GHz Citizens Broadband Radio Service (CBRS) band, to predict spectrum occupancy and interference levels. Exploratory data analysis (EDA) uncovered insights to enable early radar detection and congestion avoidance. ML models – Random Forest, SVM, XGBoost, DBSCAN, PCA, LightGBM, and regression – have been applied to address spectrum management challenges, with XGBoost excelling in occupancy prediction and Random Forest in classification. The findings support our proposed ML framework for intelligent spectrum management with implications for future (6G) networks in optimizing allocations, mitigating interference, and informing regulatory decisions.