Cybersecurity has an important and effective role in the field of information technology. Due to the increase in electronic attacks in various fields, enterprise building had taken a very important role in many researches. Many enterprises were exposed to different types of cyber attacks that change and destroy data on the network. Because IoT systems were designed with limited devices and lightweight protocols, security was a critical prerequisite for their success.
The suggested system were divided into two stages event monitoring and intruder detection based on hybrid deep learning and machine learning techniques. In first proposed, the First Hybrid of Convolution Neural Network and Long Short Term Memoery (FHCNN-LSTM) were utilized for monitoring building (event detection) in nested method without output layer to reduce the load in the Raspberry Pi. The input data of this phase was local object image from camera sensor and which number about 1400 images taken from inside and outside the building.
In second proposed, the intrusion detection system, Second Hybrid CNN and LSTM algorithm named (SHCNN-LSTM). Four datasets (UNSW-NB15, NSL-KDD, N-BaIoT and CICIoT2023) were used to train it. Also, Hybrid Machine Learning algorithms (HML) were utilized to detect attacks in building network. Different machine learning algorithms were (XGBoost, DT, RF, KNN, SVM, and NB). The output classification layer was computed based on RUF function which was a combination from one of the machine learning algorithm and set of activation function to classify the event type.
This work too presents a security system to protect features of hardware in a set of rooms from intruders and accidents inside the building based on Raspberry Pi was programmed to perform the encryption using Hybrid PRESENT SPECK algorithm (HPSPECK) to encrypt the features based on FHCNN-LSTM before send them to admin side.
The evaluation metric were presented in this work ( Accuracy, Precesion, Recall and F1 score) as performance for study. The event monitoring proposed system FHCNN-LSTM showed the best results with the highest accuracy of 99.81% of binary classification. As for the intrusion detection proposed, the highest accuracy was 99.99% for binary classification, it is attributed to the use of the RUF function, which is a set of activation functions that increase of accuracy, 99.12% for multi-classification on N-BaIoT data, 99.65% for multi-classification on CICIoT2023 data. Hybrid intrusion detection system based on HML proposed results the best highest accuracy 99.45% with CICIoT2023 dataset. Also, (HPSPECK) algorithm had achieved by NIST and lower execution time compared to the SPECK algorithm. Through these results, the proposed system was ability on detect intruders and attacks in building as well as the problems was solved and the desired goals was achieved.