In the current digital era, rapidly growing cyber threats have made cybersecurity the most concerning domain. The purpose of this study is to determine the challenges initiated by the intruders in the security of critical information infrastructures and to discover some of the diverse approaches that improve the efficiency, accuracy, and performance of intrusion detection systems. The proposed work finds out which machine and deep learning techniques are efficient for security, especially protection from web-based attacks, and it also explores the difference between the machine and deep learning techniques in improving the detection and mitigation of cyberattacks and their overall impact on cybersecurity. The primary goal is to comprehensively review and analyze the network performance in the presence of web-based attacks, DDoS, and DoS and to propose solutions that strengthen network protection using several benchmark datasets. This study presents a comprehensive analysis of various papers on web-based attack detection in cybersecurity. In a previous study, employing a lambda architecture with LSTM, artificial neural network classifiers, and convolutional neural networks achieved high accuracy with less processing time through a multi-pronged classification. Specifically, a CNN + GRU model achieved classification accuracy of 98.73% with a low false positive rate. Deep learning algorithms such as RNN types, CNN, and autoencoders, when applied to benchmark datasets including KDD99, NSL-KDD, CIC-IDS2018, TON IoT, etc., have shown improved results in terms of accuracy, precision, recall, F1 score, and availability as compared to traditional machine learning algorithms like SVM, decision trees, and RF.