Abstract:

Vehicle fuel consumption and emission have been a great deal for global warming and the world economy. The impacts of CO2 emission can be minimized through optimization of engine design parameters such as Rated horsepower(RHP), Number of Cylinders and Rotors(NCR), Number of Gears(NG), and Equivalent Test Weight(ETW) to the Rounded and Adjusted Fuel Economy (RAFE). This article explorers the weighted impact of the independent variables RHP, NCR, NG, and ETW to RAFE using Multi-Linear Regression(MLR) in machine learning. For the proposed MLR method, the vehicle data is divided into two as training and testing. Then, the data cleanup process was applied to the training data to eliminate outliers that led to incorrect predictions. The proposed method determines the correlation coefficient to compare and seek the variables having less relationships with the dependent variable RAFE. Since there are no insignificant parameters in correlation analysis, MLR training was carried out by taking into account all parameters. Finally, the processed data are trained to create a multi-linear regression model. The obtained model is evaluated through Analysis of Variance(ANOVA). According to the ANOVA, there is a significant relationship between the dependent variable RAFE and the independent variables NG, ETW, and RHP with a p-value of 4.0994e-60, 1.5887e-48, and 2.5494e-31, respectively. Moreover, p-values of NG, ETW, and RHP are supported with F-test results of 227.73, 220.87, and 152.41. On the other hand, the obtained model is also relatively less affected by NCR, with a p-value of 0.031276 and an F-test of 4.94. As a result, the resulting MLR model can be used in new vehicle designs as it reveals which vehicle parameters affect CO2 emissions.

Keywords: