This paper presents a study on improving the prediction of solar energetic particle (SEP) events using multivariate time series data augmentation. The research focuses on predicting SEP events with energy levels of approximately 30, 60, and 100 MeV, as well as non-SEP events. The study uses data augmentation techniques such as Gaussian noise, SMOTE, and ADASYN to increase the number of SEP samples, thereby improving the accuracy and F1-score of classifiers. The study covers solar cycles 22, 23, and 24, and the results show that using data augmentation methods significantly improves the performance of machine learning models, especially for the time series forest (TSF) classifier, where the average accuracy increased by 20%, reaching around 90% for the 100 MeV SEP prediction task. The study also demonstrates that using multivariate time series data of proton flux improves prediction accuracy. A comprehensive hierarchical classification framework is developed for the 30, 60, and 100 MeV SEP and non-SEP prediction scenarios. The study uses data from the Integrated Geostationary Solar Energetic Particle (GSEP) event catalog, the Heliophysics Events Knowledgebase (HEK), and GOES proton channels. The results show that data augmentation techniques significantly improve the performance of classifiers, with the Gaussian noise method being particularly effective for distinguishing between SEP and non-SEP events in the 60 and 30 MeV energy bands, while SMOTE and ADASYN outperform Gaussian for the 100 MeV energy band. The study also compares the performance of different classifiers, including ROCKET, SHAPELET, and TSF, and finds that TSF consistently achieves the best performance across all tested scenarios. The study concludes that data augmentation techniques can significantly enhance the predictive capabilities of machine learning models for SEP event prediction.This paper presents a study on improving the prediction of solar energetic particle (SEP) events using multivariate time series data augmentation. The research focuses on predicting SEP events with energy levels of approximately 30, 60, and 100 MeV, as well as non-SEP events. The study uses data augmentation techniques such as Gaussian noise, SMOTE, and ADASYN to increase the number of SEP samples, thereby improving the accuracy and F1-score of classifiers. The study covers solar cycles 22, 23, and 24, and the results show that using data augmentation methods significantly improves the performance of machine learning models, especially for the time series forest (TSF) classifier, where the average accuracy increased by 20%, reaching around 90% for the 100 MeV SEP prediction task. The study also demonstrates that using multivariate time series data of proton flux improves prediction accuracy. A comprehensive hierarchical classification framework is developed for the 30, 60, and 100 MeV SEP and non-SEP prediction scenarios. The study uses data from the Integrated Geostationary Solar Energetic Particle (GSEP) event catalog, the Heliophysics Events Knowledgebase (HEK), and GOES proton channels. The results show that data augmentation techniques significantly improve the performance of classifiers, with the Gaussian noise method being particularly effective for distinguishing between SEP and non-SEP events in the 60 and 30 MeV energy bands, while SMOTE and ADASYN outperform Gaussian for the 100 MeV energy band. The study also compares the performance of different classifiers, including ROCKET, SHAPELET, and TSF, and finds that TSF consistently achieves the best performance across all tested scenarios. The study concludes that data augmentation techniques can significantly enhance the predictive capabilities of machine learning models for SEP event prediction.