Spacecraft conjunction management plays a crucial role in the mitigation of space collisions. When a conjunction event occurs, resources and time are spent analyzing, planning, and potentially maneuvering the spacecraft. This work contributes to a subpart of the problem: Confidently identifying events that will not lead to a high collision probability, and therefore do not require further investigation. The method reduces the dimensionality of the data via principal component analysis (PCA) on a subset of features. High-risk regions are then determined by clustering the projected data, and events that do not belong to a high-risk cluster are pruned. A genetic algorithm (GA) is developed to optimize the number of clusters and feature selection of the model. Furthermore, an ensemble learning framework is proposed to combine the suboptimal models for better generalization. The results show that the first set of parameters pruned approximately 50% of the events in the testing set with no false negatives, whereas the second set of parameters pruned 70% of the events and maintained a near-perfect recall. These results could benefit the optimization of operational resources and allow operators to focus better on the events of interest.