Data Imputation Techniques Using the Bag of Functions: Addressing Variable Input Lengths and Missing Data in Time Series Decomposition

In time series analysis, the ability to effectively handle data with varying input lengths and missing data is crucial for accurate modeling. This paper presents the Bag-of-Functions-Driven Imputation framework, which leverages sequence-length independent techniques to decompose time series data while accommodating inputs of differing sizes. Unlike traditional methods that require uniform input lengths, the Padding-BoF framework employs a flexible encoding approach, allowing for the integration of variable-length time series and missing elements in the data. Through a series of experiments, we demonstrate that the BoF framework not only ensures precise reconstruction of the original data but also enhances data imputation capabilities by utilizing decomposed components. The results show that this method offers significant advantages in scenarios involving irregular sampling and disparate operational cycles, making it a valuable tool for applications in fields such as finance, healthcare, and industrial monitoring.