Cyber-attacks on the numerous parts of today’s fast developing IoT are only going to increase in frequency and severity. A reliable method for detecting malicious attacks such as botnet in the IoT environment is critical for reducing security risks on IoT devices. Numerous existing methods exist for mining IoT networks for previously discovered patterns that may be exploited to improve security. This study used a hybrid deep learning approach, namely the CNN-LSTM technique, to detect botnet attacks. Any software that infiltrates a computer system or is installed there without the administrators’ knowledge or permission is malicious. There is a wide range of viruses that cyber-criminals use to further their nefarious ends. A revolutionary deep learning system has been developed to counteract the increasing quantity of harmful programs. The system takes advantage of NLP methods as a baseline, mixes CNNs and LSTM neurons to capture local spatial correlations, and learns from successive long-term dependencies. Spatial invariance, often known as symmetry, is the property wherein the dataset size remains constant throughout iterations of an algorithm while undergoing various transformations. Therefore, automated extraction of high-level abstractions and representations aids in the malware categorization process. When compared to its predecessor research study, the current level of categorization accuracy is significantly greater than 0.81. The proposed CNN-LSTM method obtained an R2 = 99.19% in the dataset, with a correlation coefficient for the CNN-LSTM technique of R2 = 100% utilizing the provided dataset. The symmetry correlation of the CNN-LSTM, which illustrates that the CNN-LSTM method has the highest detection accuracy, at 99%, among the other malware detection methods such as the SVM and DT. The rest of classifiers had an accuracy of 98% for DT, and 95% for SVM. The accuracy of the LSTM model is 99%, the precision of the CNN-LSTM is 99%, recall is 99% and F1 score is 1.