Smartphone-based location-based services (LBS) require enhanced horizontal position accuracy with integrity. Due to the mass-market nature and compact design of smartphones, they utilize low-cost antennas and receivers, making them susceptible to multipath effects and other errors, which complicates the differentiation between reliable and unreliable measurements. To address these challenges, this paper explores the application of an adaptive Kalman filter technique to improve smartphone positioning accuracy. Adaptive Kalman filters adjust parameters such as process noise covariance or measurement noise covariance to modify the filter gain. When augmented with outlier detection mechanisms, the filter becomes more robust. This paper introduces a robust adaptive Kalman filter to enhance smartphone position accuracy. Outliers are detected using standardized innovations as a learning statistic, and a t-test is applied to these statistics to identify and mitigate outliers and adapt the measurement noise covariance accordingly. While previous research used empirical values for thresholds to adapt measurement noise covariance matrix, this study derives thresholds from t-tests, contingent on the normal distribution of learning statistics. By eliminating clock reset effects, innovations are transformed from bimodal to a normal distribution. Testing across multiple datasets demonstrates reductions of up to 42% in horizontal positioning root mean square error, with 50th, 68th, and 95th percentile statistics showing improvements of up to 53%, 41%, and 61%, respectively.