Time Series Prediction Under Data Security Based on Anonymous Features

Predicting time series remains a significant challenge in various domains. Despite advancements in big data and machine learning that have enabled sophisticated algorithms for quantitative analysis, organizations often handle sensitive information, making robust data security essential. This necessity drives the development of predictive systems that rely on anonymized features while ensuring privacy protection. Time series data is dynamic and non-stationary. Traditional models struggle with generalization due to shifts in data distribution. To tackle this issue, this study introduces a approach for time series prediction under data security constraints. By utilizing adversarial validation and anonymized features, a classification model is constructed to identify discrepancies between training and test data distributions. Subsequently, resilient anonymized features are carefully selected to reduce the impact of distributional changes on model performance while maintaining privacy safeguards. Experiments show that integrating feature selection and data preprocessing based on adversarial validation significantly enhances the accuracy and stability of predictive models, particularly in scenarios involving sudden changes in unseen patterns. This methodology not only improves predictive capabilities but also offers a reliable solution for managing anonymized data under strict data security requirements. The proposed framework provides an innovative and efficient approach to time series prediction while addressing critical concerns related to privacy and data security.