This paper introduces an effective yet simple and practical solution to improve small-cell performance in heterogeneous networks (HetNets). The proposed solution is based on deploying a switched multielement antenna (MEA) system capable of generating a variety of antenna patterns at small-cell base stations (BSs). Then, antenna patterns are assigned to user equipment (UE) in a dynamic basis. The antenna pattern selection for each UE is considered to be a supervised machine learning classification problem, in which the small-cell BS seek to find the optimal antenna pattern to serve each UE according to its measurement reports (i.e., UE radio-frequency fingerprint). Simulation results confirm the feasibility of the proposed approach, despite potential inaccuracies in UE measurement reports. Compared with the existing solutions comprising a single omnidirectional antenna (ODA), the proposed approach results in a 68% additional network-wide capacity increase. In addition, a technoeconomic analysis is presented in this paper, indicating the economic advantages of deploying the proposed scheme.