Machine Learning and Prediction of Masked Motors With Different Materials Based on Noise Analysis

Abstract

The effect of noise on the human body has attracted increasing research attention. In particular, many factories generate motor noise pollution, which exposes general workers to noise for extended periods. To solve this problem, masks made of different materials are used for reducing the noise generated by motors. In this study, we attempted to predict the acoustic sound of masked motors. We collected noise level data in decibels for different operation frequencies of motors used at National Synchrotron Radiation Research Center (NSRRC) and developed a machine learning model according to the characteristics of the collected data to simulate the effect of masks on the motor sound. We use the Gradient Boost Model (GBM) as the main learning method because the model is suitable for predicting noise from comparison results of the five models are very common predictive models and may performed as compare method to predict acoustic noise. The results indicated that the prediction accuracy of the GBM was considerably higher than other four traditional machine learning methods (random forests, support vector machine, gaussian processes regression model and multiple linear regression models). Moreover, we used a general multiple linear regression method as the worst method of comparison and conducted time-frequency visualization of the sound for analysis. At NSRRC, we examined the effects of three observation locations and three mask materials, namely wood, metal, and acrylic, on the sound prediction accuracy achieved with the developed model. The highest sound prediction accuracy was obtained behind the motor and under an acrylic mask.INDEX TERMS Gradient boosting model (GBM), machine learning, motor noise prediction, timefrequency diagrams.

Selection of Best Machine Learning Model to Predict Delay in Passenger Airlines

“…The main reason for considering the random forest algorithm for predicting flight delay can be explained with the information criteria, which works on the principle of probabilistic measure to select the best model among various machine learning models by calculating the values of various selection measures such as Akaike information criterion (AIC),Schwartz Bayesian information criterion (SBIC), and Hannan-Quinn information criterion (HQIC) [18]. Now, information criteria, i.e., AIC, SBIC, and HQIC, measures the relative information loss which is considered to select the best model based on their calculated values for the considered machine learning models.…”