With the advent of quantum computers, traditional public key cryptosystems will no longer be safe to use for the exchange of private information over the web. This shift necessitates the development of quantum-safe cryptosystems to ensure secure online communication. Out of the different hard problems secure against quantum computers, lattice-based encryption schemes, viz., the \(\textsf{LWE}\) problem, are well understood from a security perspective; furthermore, they are versatile and provide reasonable performance characteristics. The hardness of the \(\textsf{LWE}\) problem is based on learning noisy data in the presence of modulo operation. Over time, machine learning models have become better and are known to learn patterns even from noisy data samples. Recently, machine learning models have been employed to challenge the hardness of the \(\textsf{LWE}\) problem. However, the models used are resource-intensive transformer models, requiring significant computational power and time. In this work, we demonstrate that we can replace these expensive transformer models with simpler models while achieving the same success rate. Our approach uses only one-fourth of the computational resources for data preprocessing and reduces the time for secret key recovery to 1.5 min, compared to 2 h with transformer models.