A novel approach to handover management for Long-Term Evolution (LTE) femtocells is presented. Within LTE, the use of self-organizing networks (SONs) is included as standard, and handover management is one of its use cases. Base stations can autonomously decide whether handover should take place and assign the values of relevant parameters. Due to the limited range of femtocells, handover requires more delicate attention in an indoor scenario to allow for efficient and seamless handover from indoor femtocells to outdoor macrocells. As a result of the complexity of the indoor radio environment, frequent ping-pong handovers between the femtocell and macrocell layers can occur. A novel approach requiring a small amount of additional processing using neural networks is presented. A modified self-organizing map (SOM) is used to allow a femtocell to learn the locations of the indoor environment from where handover requests have occurred and, based on previous experience, decide whether to permit or prohibit these handovers. Once the regions that coincide with unnecessary handovers have been detected, the algorithm can reduce the total number of handovers that occur by up to 70% while still permitting any necessary handover requests to proceed. By reducing the number of handovers, the overall efficiency of the system will improve as the consequence of a reduction in associated but unnecessary signaling. Using machine learning for this task complies with the plug-and-play functionality required from SONs in LTE systems.