Volume 15, Number 1, June 2025
| Forecasting Data-Driven Performance: Learning-based Data Manipulation Detection System
|
Cheng-Shian Lin 1*, Sin-Man Wong 2, Chien-Chang Chenh 3
Abstract
Malicious data tampering detection is of importance in protecting the integrity and reliability of data-driven systems, such as the disaster warning system. In this study, we proposed a learning-based data manipulation detection system. The proposed system that integrates long short-term memory network (LSTM) and autoencoder models, in which the LSTM neural network is utilized serves as the encoder to capture the feature of time and condense input time series into lower-dimensional data. Subsequently, a decoder employs LSTM neural network to reconstruct this reduced-dimensional data to its original time series data. Finally, the reconstructed data caused by the anomy errors is detected as malicious tampering by our proposed confusion-based optimal thresholding method. Experimental results show that the proposed approach outperforms the conventional models, and can efficiently detect and localize various malicious tampering methods.
Keywords: Long short term memory (LSTM); autoencoder network; data manipulation detection
JEL Classification: C53, C45, M42
1 Department of Computer Science and Information Engineering, Tamkang University
2 Department of Computer Science and Information Engineering, Tamkang University
3 Department of Computer Science and Information Engineering, Tamkang University