We present an evaluation of our neural network pipeline applied to gravitational microlensing detection and event classification. Our generalized photometric neural network pipeline automatically identifies and characterizes events in photometric light curves. The pipeline uses the raw light curves and does not require any other prior modeling or information on the light curve. Using the flux of the light curve without pre-extracting extra features can have a few advantages. First, the approach has no intrinsic biases. Second, using only the network is very fast compared to typical detection methods. Third, since neural networks can automatically learn complex data features, it has the potential to outperform hand-crafted methods. In previous work, we have successfully applied it to identify exoplanet transit candidates using light curve data from the Transiting Exoplanet Survey Satellite (Olmschenk et al. 2021), which indicates that a similar approach applied to gravitational microlensing is very promising. Now, we present the results of the pipeline when trained with 549,447 previously classified light curves acquired by the Microlensing Observations in Astrophysics (MOA collaboration during nine years, from 2006 to 2014. The MOA collaboration conducted the first high cadence microlensing surveys towards the Galactic bulge and has previously reported the planet frequency as a function of planet-to-star mass ratio (Suzuki et al. 2016). The Suzuki et al (2016) analysis defined the mass-ratio threshold \(q < 0.03\) as the upper limit for planetary events because it was comprised of events that were detected as single-lens events by the MOA alert system. We expect to use this pipeline to determine the detection efficiency within the MOA data set for all mass ratios, through the brown dwarf desert to stellar binaries, which is required for the statistical understanding of the exoplanet and binary star distribution. We will also evaluate our pipeline as an alternative detection method.