Smart sensor-based system for campus fire detection and control using a Naive Bayes classifier

https://doi.org/10.36386/rj6n8d10

Emergency incidents relating to fire control and recovery systems require urgent attention to detect fire explosions quickly and to speed up the control process in order to allow escape. In gigantic buildings and large-scale environments like a university campus, data transmission to the control panel for signal processing may be obstructed; the inability to provide accurate analytical sensor data could cause excessive false alerts. Security monitoring of campus to protect lives and property through immediate response to pre-empt fire outbreak is the objective of this study. This study adopts a quantitative and experimental design by incorporating smart sensor being integrated into portable devices like personal digital assistants (PDA) or internet enabled smart phone; with the use of supervisory trained Naive Bayes classifier for intelligent reasoning and decision-making through possibility approximation analysis of environmental observations as a machine learning model for fire detection. An improved model (Fire-DC) for implementing a fire detection and control system in a campus environment was developed using smartphone sensors and a Naive Bayes classifier. Seventy-five percent (75%) of the dataset was provided as sensor values and experimental observations to the model (Fire-DC) for training, while twenty-five percent (25%) of the dataset was later used for testing in scientific experimentation. Performance of the improved system (Fire-DC) was evaluated through a web-based application program implemented in HTML, Bootstrap, XML, PHP, MySQL, Google Maps API, and integrated sensors in internet-enabled smartphones, and compared with previous techniques in existing systems using accuracy, sensitivity, precision, and confusion matrix as yardsticks. Prototype development and experimental performance of the improved system (Fire-DC) recorded higher accuracy of ninety percent (90%) for its classification algorithm; as well as good sensitivity of seventy one percent (71%) for smart sensors in sophisticated mobile phones to detect presence of fire in immediate surroundings, with visual description of incidence location through geographic positioning system (GPS) coordinates being fetched from sensor nodes.