Heterogeneous networks (HetNets) have been proposed as a capacity and coverage enabler in LTE-Advanced and beyond communication networks. Their optimal operation requires a significant degree of self-organization. Autonomic Load Balancing (ALB) has been proposed as an important self-organizing (SON) function in the LTE radio access network (RAN). In this work, distributed ALB is achieved by implementing a programmable autonomous learning model. The optimization problem (load balancing) is split into many small optimization problems and tasks, which are solved by using machine learning algorithms. The load conditions of the E-UTRAN NodeB (eNBs) and the measurement reports from the mobile terminals are used for creating a decision map for the load balancing. The simulation results show that by using ALB, the system capacity can be improved significantly.