The albedo of near-Earth asteroids (NEAs) remains poorly known due to observational limitations, with albedo values for 96% of known NEAs currently undetermined. Jet Propulsion Laboratory uses a range from 0.05 to 0.25 as estimation, which can lead to significant uncertainties. In this work, we hypothesize that NEAs originated from unstable regions of the asteroid main-belt, and that their albedo distributions are statistically correlated with those of their source regions. We propose a novel method to estimate the albedo of NEAs based on a neural network and a statistical method. First, a supervised learning method is used to identify main-belt family memberships, assuming that members of the same family share similar albedo properties. Second, the derived albedo values of main-belt asteroids from WISE/NEOWISE are used to train an artificial neural network model, taking the orbital elements and the family information as inputs. This model achieves a 6.5% improvement in predictive accuracy compared to the approach proposed by Murray. We characterize the distribution of albedo for each source region via the kernel density estimation method. Then, we calculate joint distribution and corrections for space weathering effects to estimate the distribution of albedo for an individual NEA. Defining the cross-entropy below 0.5 as a criterion, for a sample of 1110 NEAs with known albedo, over 98% of the estimations can be considered valid.