With fast growing data collected by the Fermi Large Area Telescope as a big data problem, manual classification has become an impossible task for astronomers. In this paper, we propose a novel framework using machine learning techniques for gamma-ray object classification. We use the random forest (RF) algorithm for feature selection in order to achieve better classification performance. After an extensive experimental study with feature selection incorporated, we found the best results can be obtained for both active galactic nuclei (AGN) / pulsars (PSR) classification, and young (YNG) / millisecond pulsars (MSP) classification using boosted logistic regression (LR). We automate parameter tuning rather than manual tuning used in previous works. We compare the performance based on our framework with those based on Saz Parkinson et al. (2016) \cite{5} by using the data obtained from the 3$^{\rm rd}$ Fermi Large Area Telescope Source Catalog (3FGL) \cite{F3}. In PSR/AGN classification, we achieve an accuracy of $>98\%$. On the other hand, we attain an accuracy of $>95\%$ in the case of YNG/MSP classification.