In this project, to better predict finger flexion for individual subjects, we developed an algorithm called logistic-weighted random forest, which utilized the classification results from logistic regression as a weight, and then combines the nonlinear regression results from random forest to generate a final data glove prediction. LASSO was incorporated in the logistic regression to allow variable selection and regularization. The addition of random forest is necessary in that it can provide non-linear prediction results required by the problem, while avoid overfitting commonly occurred in polynomial regression. The entire algorithm includes data pre-processing, feature extraction, feature processing, model training, prediction, cross-validation, and post-processing. This pipeline achieved 52.8% and 46.7% correlation on the public leaderboard and private leaderboard, listed as Top 5 solution in the 2020 BE521 Brain-Computer Interface Final Competition.