In this study, machine learning (ML) based methods are used to estimate rotor mechanical speed of brushless direct current (BLDC) motors. Training performances of approaches such as Artificial Neural Network, k-Nearest Neighbor, and Random Forest in the ML-based speed estimator are tested using the datas obtained from the direct torque control (DTC) drive system of BLDC motor in simulation and it is seen that the ANN approach has the highest accuracy. In addition, a novel extended Kalman filter (EKF)-based estimator is proposed for the estimation of back-EMFs of BLDC motor. A hybrid estimation method is proposed by using the developed ML-based speed estimator with the proposed EKF-based estimator and its estimation performance is tested in simulation on DTC drive system.