This study investigates the police treatment of individuals arrested for the possession of small quantities of marijuana in Toronto. Utilizing a dataset comprising 5226 observations, we analyze the factors that influence whether an arrestee is released with a summons. The dataset includes variables such as the arrestee's race, age, sex, employment status, citizenship status, the year of the arrest, and the number of police databases in which the arrestee's name appeared. A neural network model is developed to predict the likelihood of release based on these factors. Our findings indicate significant patterns and disparities, shedding light on the influence of demographic and socioeconomic factors on police decision-making. The results underscore the potential of machine learning models in uncovering biases and guiding policy reforms in the criminal justice system.