Accurate channel estimation is a cornerstone for ensuring the performance of Large Intelligent Surface (LIS) systems, where the reliability of communication relies heavily on precise knowledge of the channel matrix. Conventional Least Squares (LS) estimation, while simple, suffers from high sensitivity to noise and poor performance in ill-conditioned scenarios, limiting its effectiveness in LIS environments. This paper proposes replacing LS with Regularized Recursive Least Squares (RLS) for channel estimation in LIS systems. The Regularized RLS method improves estimation accuracy by incorporating a regularization term to stabilize the solution in the presence of noise or poorly conditioned pilot matrices. Through theoretical analysis and simulation results, we demonstrate that Regularized RLS achieves better Bit Error Rate (BER) performance as a function of \(E_{b} /N_{0}\) compared to LS, while maintaining computational efficiency. Unlike conventional RLS, the proposed method adds a regularization term that significantly improves robustness in low SNR and ill-conditioned environments, which are common in LIS systems. Furthermore, by integrating Low-density parity-check (LDPC) coding, the system achieves even lower BER, highlighting its effectiveness in mitigating errors and improving communication reliability. The proposed method is particularly well-suited for large-scale LIS systems, providing a robust and practical alternative to LS.