Classification of measured chemical gases by machine learning

In this thesis, the purpose of this study is to classify the sensor data obtained by using embedded system software and hardware, which is one of the fundamental issues of industry 4.0, with machine learning algorithms and to create the infrastructure for the necessary automation for sustainable production. Today, all devices used in the industry become smart and digital data produced at all levels of the production process is used to increase product quality, flexibility and functionality. Data from production facilities are collected with sensors. It is not possible for human data to be evaluated by these data, which are mentioned with the definition of big data. It is only possible to evaluate this data with machine learning algorithms, to make sense of it and to use it for sustainable production. In this study, the perception and classification of gases are emphasized in the industry both because it directly affects the human life and because it occurs or uses many products. Gas sensors are widely used in industry and fire fighting to detect flammable, combustible and toxic gases that threaten human health and properties. Firstly, it was studied on collecting b gas data with Ardunia and classifying it with machine learning. Since sufficient data about gases could not be obtained, a dataset dataset was downloaded for six different volatile organic gases and studies were carried out on it. In this dataset; There are ammonia, acetaldehyde, acetone, ethylene, ethanol and toluene gases. It was obtained over a three-year period using 16 metal oxide gas sensors. These data contain 13910 measurements and 129 features. In our study, many machine learning algorithms were used to classify this data. These; k-nearest neighbors, Support vector machine, Random forest and Logistic regression stop. These machine learning algorithms were implemented in the Weka program to classify the data. As a result of the studies, the k-nearest neighbors algorithm gives the best result and the classification success is 99,48%.