Abstract—This paper introduces a hybrid, low-complexity channel estimation scheme that integrates pilot-aided channel estimation with Machine Learning (ML) to obtain the Channel State Information (CSI) for an Orthogonal Frequency Division Multiplexing (OFDM) system. The proposed method eliminates the need for channel statistics at data subcarrier positions, while still achieving performance comparable to that of classical pilot-aided channel estimation schemes. The CSI is fully determined by first obtaining channel coefficients at subcarrier positions where pilot symbols are transmitted; these coefficients are then used to obtain channel coefficients at data subcarrier positions. Thus, in the proposed method, pilot symbols are inserted among data symbols. The pilot symbols are used to estimate channel coefficients at pilot subcarrier positions using the Linear Minimum-Mean-Square-Error (LMMSE) estimator. The estimated coefficients are then used as an input to a Feed- Forward Neural Network (FFNN). The FFNN is trained to learn and capture channel characteristics from the estimated coefficients at subcarrier positions where pilot symbols are transmitted. The trained neural network can predict channel coefficients at data subcarrier positions where pilot information is unavailable. In some conventional pilot-aided channel estimation schemes, the cross-correlation matrix of channel coefficients at the positions of data subcarriers is an essential parameter for estimating channel coefficients at those positions. However, this statistical information may not be available. In the proposed scheme, channel coefficients at data subcarrier positions can be predicted using a trained FFNN, without requiring statistical information about coefficients at those positions. Also, the proposed method avoids high computational complexity in statistical-based channel estimation methods. For example, the computational complexity of the classical LMMSE mainly stems from the matrix inversion and multiplication operations it involves. Computer simulations in MATLAB show that the Mean- Squared Error (MSE) of the proposed estimation method can achieve performance comparable to classical systems, especially at low Signal-to-Noise Ratio (SNR).