Due to the complex mechanism of the effect of micro-source switching on interactive stability between the grid and the micro-grid with series micro-source inverters, traditional methods were difficult to model and can't make decision in time. In this study, therefore, an assessment method of interactive stability based on gradient boosting decision tree (GBDT) was proposed. Firstly, the feature functions of the micro-source switching were constructed to extract the position feature which existed typically in series structure system modulated by carrier phase shifted SPWM. The proposed functions can reduce the feature dimension, thus improving the efficiency of assessment model and making the model migration capable by feature fitting. Secondly, the assessment of interactive stability was treated as a multi-classification task and the resonant frequency bands of instable samples were predicted. Thus, the different strategy can be taken according to the classification and prediction. Finally, the tests on validation data sets and new test sets verified the correctness and effectiveness of the feature construction functions which can provide references for the similar series structure system in power quality analysis, optimal operation and other aspects. The results also verified the high generalization performance and practicality of the proposed assessment model, which can provide a new tool for on-line stability analysis and decision-making of grid-connected operation systems.