We present an analysis of the long-term performance of the W. M. Keck observatory laser guide star adaptive optics (LGS-AO) system and explore factors that influence the overall AO performance most strongly. Astronomical surveys can take years or decades to finish, so it is worthwhile to characterize the AO performance on such timescales in order to better understand future results. The Keck telescopes have two of the longest-running LGS-AO systems in use today, and as such they represent an excellent test-bed for processing large amounts of AO data. We use a Keck-II near infrared camera 2 (NIRC2) LGSAO surve of the Galactic Center (GC) from 2005 to 2019 for our analysis, combining image metrics with AO telemetry files, multiaperture scintillation sense/differential imaging motion monitor turbulence profiles, seeing information, weather data, and temperature readings in a compiled dataset to highlight areas of potential performance improvement. We find that image quality trends downward over time, despite multiple improvements made to Keck-II and its AO system, resulting in a 9 mas increase in the average full width at half maximum (FWHM) and a 3% decrease in the average Strehl ratio over the course of the survey. Image quality also trends upward with ambient temperature, possibly indicating the presence of uncorrected turbulence in the beam path. Using nine basic features from our dataset, we train a simple machine learning (ML) algorithm to predict the delivered image quality of NIRC2 given current atmospheric conditions, which could eventually be used for real-time observation planning and exposure time adjustments. A random forest algorithm trained on this data can predict the Strehl ratio of an image to within 18% and the FWHM to within 7%, which is a solid baseline for future applications involving more advanced ML techniques. The assembled dataset and coding tools are released to the public as a resource for testing new predictive control and point spread function-reconstruction algorithms.