Description:The growing use of IoT technology in smart offices has increased the potential of network assaults and security issues. Ensuring exceptional safety in smart offices systems is critical for protecting confidentiality and data. While these developments provide tremendous advantages, they also introduce new obstacles. If these smart settings are hacked by attackers, major privacy breaches may occur, resulting in data theft or activity monitoring. As a result, an encrypted IoT framework for anomaly detection is required to enable safe communication among sensing devices. Smart gadgets can increase people's vulnerability to security threats. Generally, for a company, agency or officer, the data is critical to their operations. IoT nodes can provide hackers access to sensitive information from any large organization. These challenges are frequently handled using relatively easy techniques. A signature-based method is often used to detect abnormalities in advance, and the system is occasionally cross-checked against a database. However, this strategy increases processing costs and exposes the system to unforeseen attackers. Besides of this, the technique for selecting the appropriate features is the key for determining the most important qualities. To deal with unknown assaults, a data-driven technique proven to be safer than alternative approaches. As a result, this study employs data-driven approaches.
ML has altered smart office settings by allowing IoT gadgets to interpret data with supervised machine learning algorithms. The proposed model analyzes data from various sensors and actuators using a supervised learning algorithm. This strategy involves training the algorithm on labeled data before making predictions based on that training. Unsupervised learning, on the other hand, eliminates the need for prior data knowledge. In this step, labeled training data is utilized to train the classifier for learning and classification tasks. The model's performance is then evaluated by testing with unlabeled data. Despite current research, security and threat prevention for smart workplaces have not yet been completely created to match domain needs. To maintain a safe smart office, these settings require improved security mechanisms, privacy protections, and data transfer protocols, all of which are enhanced by machine intelligence. To address these security concerns, our proposal presents a machine learning-based layered architecture that provides improved and reliable anomaly detection in smart offices.
This study uses the following steps to perform the task:
1) Collection of datasets: First, the dataset from various IoT devices installed in the office is generated and collected.
2) Preprocessing of dataset: The data collected in the step 1 is then pre-processed. This step involves the data quality check like finding missing values, data balancing, wrangling, visualization, normalization, encoding etc.
3) Selection of Features: This step is to selecting the relevant features to the target variable. Pearson correlation is one of the most useful techniques to find the perfect features. This technique also reduces the excess and irrelevant features from the dataset.
4) Data stratifications: In this stage, the data is stratified in to three sub data: training, validation and testing dataset. Generally, 60:20:20 ratio is selected to split the dataset inton training: validation: testing.
5) Hyperparameter selection: This proposed work utilized the BiLSTM (Bidirectional Long short-term memory) to detect the anomaly in the devices. This is the deep Learning (DL) model used for regression and classification task. However, the accuracy of the model depends upon the fine tuning of hyperparameters of the model. Therefore, the training and validation dataset is used to find the optimal hyperparameters of the DL model.
6) Evaluation of the model: This study utilized the classification task by the BiLSTM DL model to detect the anomalies. To evaluate the performance of the model, different parameters like accuracy, sensitivity, precision, confusion matrix is used.
, C , Claims:1. Our created model utilizes an effective and advanced Bidirectional Long Short Term Memory Deep learning model to detect the anomalies of the IoT devices.
2. We claim that the developed model not only useful for a single IoT domain for security systems but also useful for detecting the different anomalies of the IoT systems
3. We claim that the developed model definitely useful for the upcoming days with the increase in installation of IoT devices. The security of database in the device will be major concern. This approach will be revolutionary approach in the Industrial IoT market.
4. We claim that by this system accurate and precisely detect the anomaly related to the IoT devices.