Phishing is still a big danger to cybersecurity, and it keeps changing on email, internet, SMS, and social media. This research systematically evaluates data mining (DM) methodologies for phishing detection in accordance with PRISMA 2020 recommendations. After going through 612 records from six databases, we discovered 65 studies that met our criteria. We look at several algorithms, datasets, feature engineering methods, and ways to measure performance. Ensemble models and Transformer-based designs regularly surpass conventional classifiers. Nonetheless, trade-offs among accuracy, interpretability, and resource efficiency continue to be paramount. This article outlines practical deployment issues, identifies research gaps, and suggests future possibilities, serving as a thorough reference for cybersecurity academics and practitioners.