A comparative study on appliance recognition with power parameters by using machine learning algorithms

Abstract

Recently, machine Learning algorithms are widely used in many fields. Especially, they are reallygood to create prediction models for problems which are not easy to solve with conventionalprogramming techniques. Although, there are many different kinds of machine learningalgorithms, results of applications are varying depend on type of data and correlation ofinformation. In this study, different machine learning algorithms have been used for appliancerecognition. The measurement data of Appliance Consumption Signatures database and somederivative values have been used for training and testing. Additionally, a data pre-processingtechnique and its effects on results have been presented. Filtering corrupted data and removinguncertain measurement value has improved the quality of machine learning. Combination ofmachine learning algorithms is best way to work with uncertain values. Different feature extractionmethods and data pre-processing techniques are crucial in machine learning. Therefore, this studyaims to develop a high accurate appliance recognition technique by combining grey relationalanalysis and an ensemble classification method. The results of this new method have beenpresented comparatively to show the improvement for itself and previous studies that uses thesame database.