<p>Swipe gestures, captured by the touchscreen and sometimes complemented with inertial sensing, provide a practical basis for unobtrusive continuous authentication on mobile devices. However, research on Swipe Gesture-Based Continuous Mobile Device Authentication (SG-CMDA) is scattered across datasets, protocols, and modeling choices, which complicates comparative assessment. We conducted a systematic literature review (SLR) of studies published between 2018 and 2024. From 110 records, 22 primary studies were retained after dual-reviewer full-text screening. We synthesize a reference SG-CMDA workflow–from data acquisition and preprocessing to representation learning, user modeling, and decision-making–and summarize seven commonly used public datasets: CEP, HMOG, BioIdent, BrainRun, BB-MAS, Touchalytics, and JSS Touch. Across the reviewed literature, two representation trends dominate: (i) handcrafted touch-dynamics features, and (ii) image-based encodings analyzed with CNNs. Reported performance spans a wide range, with feature-based approaches achieving 0.55%−6.35% EER and image-based methods typically reporting 4.1%−4.6% EER; recent Transformer-based models report 3.6% EER. A consistent theme is sensitivity to evaluation design and operating conditions: results often degrade substantially under cross-session testing, and performance varies even on the same dataset due to differences in preprocessing and data splits. Context changes (e.g., sitting versus walking) further reduce reliability, while two-class formulations remain constrained by the practical difficulty of collecting representative impostor data. Based on these findings, we outline research directions centered on privacy-preserving adaptation (e.g., federated learning), context-aware modeling, and standardized benchmarking protocols, with emerging bio-inspired and quantum-influenced learning as a longer-term direction for improving robustness in dynamic environments.</p>