Sudoku, a popular logic-based number placement game, presents a complex challenge that requires systematic pattern recognition and logical deduction. In this study, we explore the efficacy of diverse machine learning techniques in automating Sudoku puzzle solutions. We employ algorithms like convolutional neural networks (CNNs), dense neural networks (DNNs), long short-term memory (LSTM) model, and one customized model which is a simple neural network (SNN) model to approach the Sudoku-solving problem. Our experimental results highlight the strengths and limitations of each method regarding accuracy, efficiency, and adaptability to varying puzzle complexities. Through a comparative analysis of these techniques, we aim to contribute insights into the most effective strategies in machine learning to address Sudoku puzzles, shedding light on their potential broader applications in problem-solving domains. We have used standard Sudoku of size 9 \(\times \) 9 for our model testing and validation. After analysing all three algorithms, we can conclude that the SNN model produces quite a good accuracy compared to CNN, DNN, and LSTM for solving the Sudoku puzzle. Various approaches like backtracking, evolutionary algorithms, and permutation tree-based approaches are used as Sudoku solvers. This work highlights how the machine can be trained effectively using some popular machine learning algorithms so that it can also solve Sudoku instances in an efficient way.