Data Exploratory Analysis and Feature Selection of Low-Speed Wind Tunnel Data for Predicting Force and Moment of Aircraft

Abstract

This paper discusses exploratory data analysis (EDA) and feature selection of aircraft test results in Indonesia's low-speed wind tunnels (ILST). First, we briefly explain input and output parameters and data processing to make readable and higher accurate data. Then, we used feature selection using embedded and random forest methods to find parameters that most affect the force coefficient of aircraft. The research activities carried out in this study are to review literature from either scientific journals, the internet, or books and interview with an engineer who tests aircraft models at ILST. Then create a program for processing data from test results, such as data extraction, data cleaning, exploratory data analysis, and feature selection with python. After applying the feature selection method, we found that all the methods show similar results and have succeeded in separating the powerful features from the weak ones with a significant score difference. We decide to use the Random Forest method. The three most strongest features in the coefficient of an aircraft model in the ILST test (C_L, C_D, C_M25, C_YAW, C_ROLL and C_Y) are the following: for C_L are ALFA (0.984), T0 (0.008), P0 (0.004), on for C_D is are ALFA (0.965), T0 (0.009), RE (0.007), in C_M25 are ALFA (0.416), P0 (0.285), T0 (0.168), in C_YAW are BETA (0.44), T0 (0.141), ALFA (0.141), in C_ROLL is BETA (0.79), ALFA (0.091), P0 (0.036), and in C_Y are BETA (0.842), ALFA (0.114) and T0 (0.014). The results of this paper can be used to help build a model for the coefficient of aircraft design using machine learning based on the data from the ILST test more effectively and efficiently.