Selection of Vape Sensing Features in IoT-Based Gas Monitoring with Feature Importance Techniques

Current smoke detection sensors are not designed to specifically detect vape smoke, resulting in ambiguity in decision making. Vape and cigarette smoke are different from each other, cigarette smoke is the result of burning from tobacco, while vape smoke comes from heating liquids that produce steam in the form of smoke. Therefore, researches have been directed to design a smart device that can detect vape smoke by combining multiple gas sensors, for example a MQ2 gas sensor, a MQ7 gas sensor, and a temperature and humidity sensor connected to the microcontroller. Previous tests have returned bad accuracy in detecting Vape. To evaluate that problem, this research proposes feature importance technique to select the appropriate sensor to detect vape smoke by providing a value for each sensor according to the effect on the outcome of the decision. For classification, a random forests method is used that matches the feature importance and firebase as storage media. By using these devices and methods, the sensor is suitable to detect vape smoke so that it gets an accuracy of 94.44%.