The fundamental concept of prognostics and health management (PHM) within the scope of Condition-Based Maintenance (CBM) is to find an approach to evaluate the system health and predict its remaining useful life (RUL). Many methods and algorithms have been proposed for PHM modeling, most of which have been proven to perform relatively well. One of the leading algorithms in the current data-driven technology era is a deep learning approach, which is based on the concept of multiple hidden layers in a neural network. RUL prediction is an important part of PHM, which is the science that is aimed at increasing the reliability of the system and, in turn, reducing the maintenance cost and potential failure. The majority of the PHM models proposed during the past few years have shown a significant increase in the number systems that are data-driven. While more complex data-driven models are often associated with higher accuracy, there is a corresponding need to reduce model complexity. One possible approach is to reduce the complexity of the model is to use the features (attributes or variables) selection and dimensionality reduction methods before the model training process. In this work, the effectiveness of multiple search-based methods that seek for the best features set to perform model training, which included, filter and wrapper feature selection methods (correlation analysis, relief forward/backward selection, and others), along with Principal Component Analysis (PCA) as a dimensionality reduction method, was investigated. A basic algorithm of deep learning, Feedforward Artificial Neural Network (FFNN), was used as a benchmark modeling algorithm. It is believed that all of those approaches can also be applied to the prognostics of an aircraft engine. The aircraft engine data from NASA Ames prognostics data repository was used to test the effectiveness of the filter and wrapper feature selection methods. The findings show that applying feature selection methods helps to improve overall model accuracy by 3% to 5% compared to other existing works and significantly reduces the complexity by using 7 out of 21 less input nodes for the deep learning type of models.