<p>This paper introduces new data fusion techniques to enhance the accuracy and adaptability of drone tracking systems in complex and dynamic scenarios. The paper investigates the application of three approaches that rely on long short-term memory (LSTM) in addition to the covariance matrix adaptation evolution strategy (CMA-ES) algorithm approach. All are used to predict the optimal fusion of tracks for drone tracking data tasks. In particular, and in the first approach, the LSTM capability in capturing the mixing parameter temporal patterns is used to predict future values. In the second approach, which is the dynamic fusing approach, the LSTM learns to return fused tracks by learning how to apply fusion by itself. Furthermore, in the third approach, the fusion is achieved by incorporating the minimization of the drone tracking innovation values. The fourth approach, the CMA-ES, is used offline to build a knowledge base that is used in online inference for the fusion mixing parameter values. In all methods, the extended Kalman filter method is used for estimating the tracks. The paper comprehensively analyzes the proposed methods across multiple simulation scenarios and compares their performance against baseline approaches. The results show that the proposed methods significantly improve the accuracy and robustness of drone tracking systems in a Stone Soup simulated environment. Overall, this paper contributes to the advancement of drone tracking systems by proposing effective methods for finding the optimum mixing parameters in data fusion. Comparisons are made between the different proposed techniques based on standard metric techniques such as OSPA (optimal subpattern assignment) and the SIAP (single integrated air picture). Each method showed strengths and weaknesses; however, the LSTM second and third approaches showed advantages over the other techniques.</p>