Building wireless sensor network: prototype datalogging enhancement & development as a real-time modelling device

Abstract

This paper delves into an extensive exploration of fire sensing methodologies, placing a specific emphasis on the development of an enhanced prototype model designed for efficient dataset collection and backtesting within the framework of a "No-Fire detector" system. Subsequently, a comprehensive analysis is conducted, comparing the performance of Ordinary Least Squares (OLS), Polynomial Regression (PR) & K-Nearest Neighbors (KNN) models using a 22-day dataset of pressure data. Key performance metrics, including R2, Cross Validation Score Mean (CVSM), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE), are employed to evaluate accuracy and stability over time. The findings underscore the consistent and superior performance of the KNN model in contrast to OLS & PR. Notably, the KNN algorithm demonstrates computational efficiency, achieving a 95% accuracy within 9 minutes. This study extends to diverse locations, providing insights into the KNN model's behavior under varying environmental conditions. Overall, this research significantly contributes to advancing the development of an efficient and reliable "No-Fire detector" system.