There are diverse choices available for feature selection techniques that can be used to apply in an ML model that predicts whether the URL provided as an input is safe or not. This paper examines the effects of four main approaches present at this point in time: Recursive Feature Elimination (RFE), Mutual Information (MI), Correlation Feature Selection, and Manually Done Feature Selection. Numerous features were considered to discriminate between malware and benign URLs such as IP properties, hostname length, and abnormal URL. We have also used three different machine learning classifiers, i.e., Random Forest, XGBoost, and LightGBM to assess the performance of feature selection with different ML classifiers. The results have been assessed by calculating precision, recall, F1-score, and accuracy. This paper demonstrates why the selection of a suitable feature selection technique is critical and impacts the output generated and which decision-making system is the most fitting.