The work investigates the mitigation of adversarial attacks in malware detection systems through various regularization techniques. While ML plays an increasingly vital role in cybersecurity, it also becomes important to make the defense mechanisms robust against complex cyber threats. This presented study puts its focus on strengthening malware detection models by adopting three different regularization methods on a simple LCSGD. It regularizes the randomness and uses a logistic loss function to improve the model's performance and make the results of different runs more consistent. Results will prove that regularization can indeed manipulate decision boundaries in such a way that it becomes hard for adversaries to find vulnerabilities in malware detection systems. This work contributes to the ongoing efforts to strengthen cybersecurity frameworks against evolving threats.