In multi-institutional neuroimaging-based schizophrenia classification, inter-institutional distributional differences can adversely affect the generalization performance of machine learning models. To address this issue, we propose a Balanced ComBat pipeline that combines propensity score matching with undersampling. To ensure independence of the test data, ComBat was trained using data excluding the test center during the training phase, and applied to the full dataset during the test phase. While the application of ComBat alone reduced external AUC to 0.60, the proposed Balanced ComBat approach recovered performance to around 0.70, demonstrating the effectiveness of applying data adjustment techniques to Balanced ComBat.