Data gathering and utilization across various areas have undergone a paradigm shift with the introduction of the Internet of Things (IoT). However, new ways of data analysis are required due to the inherent constraints presented by the large scale and heterogeneity of IoT systems, as standard centralized methods are insufficient. To overcome these obstacles, this study presents a novel framework called Federated Virtual Sensors for IoT (FVSI). The FVSI framework enables distributed intelligence across IoT devices by collecting data from physical sensors and generating simulated virtual data for local model training. The server distributes the global model, created by aggregating model updates from all sensors via Federated Averaging (FA), back to the individual devices for ongoing enhancement while maintaining data privacy. This paper explains the architecture and deployment of a federated virtual sensor system that gathers data from IoT devices, simulates virtual data to enhance the dataset and enables dispersed intelligence. Using the FA communication algorithm, the solution protects data confidentiality and privacy while facilitating cooperative training of the Linear Regression(LR), Machine Learning (ML) model on local data. Within the scope of FVSI, regression algorithms such as Decision Tree (DT) Regressor, Random Forest (RF) Regressor, Support Vector Machine (SVM), Gradient Boosting Regressor, and K-nearest Neighbor (KNN) Regressor are used to evaluate the performance of the framework using three metrics Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R \(^2\) ).