Eliminating the connectivity problem with penalized regression methods

With the developing technology, interest in artificial intelligence applications has increased and has become the center of attention of institutions and academic studies. Although decision trees and artificial neural network algorithms are frequently used methods in machine learning, regression models are still among the most commonly used methods due to their suitability for the purpose of the study or the data sets used. However, in some regression models, there may be a linear or near-linear relationship between two or more of the independent variables, which is called the "Multicollinearity Problem". In cases where multicollinearity is encountered, alternative methods such as Lasso Regression and Ridge Regression can be considered. This thesis uses a dataset of 1000 records of students' grade performance, which is available as open source in the Kaggle database. Linear Regression, Lasso Regression and Ridge Regression machine learning models are applied to the dataset using Python 3.8.5 software language. As a result, in this study, the effect of penalized regression methods on supervised machine learning is tested on an example and the results are discussed. The performance values of the models applied separately on the system were realized as "0.839" in Linear Regression, "0.843" in Lasso Regression and "0.846" in Ridge Regression.