As GNSS is used in various applications that require high accuracy and high reliability, various carrier phase-based augmentation systems have been developed. Carrier phase-based GNSS Augmentation systems such as real time kinematics (RTK) and precise point positioning-real time kinematic (PPP-RTK) are able to provide centimeter-level GNSS positioning performance, and so providing reliable corrections from the reference stations is crucial. Therefore, to obtain safe and accurate PNT solutions, the augmentation systems need to assess the quality of the measurements and detect anomalies such as cycle slips prior to estimate corrections or perform integrity monitoring. To monitor the validity of the carrier phase measurement, modeling the residual under a normal condition should be preceded so that thresholds for the nominal expected residual can be defined to monitor outliers. Although there exist numerous methods to model and monitor the measurements, research on the carrier phase measurements is not as extensive as the code measurements, and many of them have limitations that do not reflect site-dependent residual anomalies. This paper proposes a method to use machine Learning based non-linear regression to effectively model and monitor potential faults in the GNSS measurements including carrier phase. This method can be divided into three steps. In the first step, the Short-Baseline Double Difference (SBDD)-based residuals are utilized to extract the site-dependent error terms. The precise and accurate relative vector between two GNSS antennas of a reference site can be used to calculate the residuals instead of using their absolute positions. The ambiguity integers are fixed throughout the pass-over by dividing all the double-difference terms in the outlier-free session by the wavelength and rounding to the nearest integer. In the second step the standard deviations of the residuals are modeled using Machine Learning based non-linear regression. The residuals obtained from SBDD consist of mainly noise and multipath components. The noise depends on satellite elevation and C/No and the multipath depends on site visibility, so these terms must be separated prior to modeling. Therefore, we separate the residuals to noise and multipath using high-pass filter and low-pass filter. In the case of high pass filtering residuals, we used a Gaussian SVM as the machine learning model because thermal noise has a strong Gaussian property. And fine tree is used for the model type for the low pass filtering residuals as multipath is discrete and environmently dependent. The inputs for the machine learning are the normalized East, North, Up (ENU) of the satellites with respect to the receivers indicates and the satellite C/No. In the third step the over-bounding sigma is calculated. In order to be employed to a safety-critical system by ensuring integrity, inflation of the nominal standard deviation of the normal residual errors with respect to the tail portion needs to be considered. The inflation factors corresponding to a real augmentation system would increase the feasibility of the real implementation.