MIRA-IOT: A Missingness-Informed Framework for Adaptive Imputation in IoT Sensor Networks

Missing data are a persistent problem in IoT sensor networks, but the difficulty of reconstruction is not determined by the amount of missing data alone. Short isolated gaps, long outages, unstable signals, and loss across related sensors create different recovery conditions. Yet many IoT pipelines still apply one imputation method to all cases. This paper proposes MIRA-IoT, a missingness-informed framework that makes imputation selection adaptive to the current data condition. MIRA-IoT first builds a Missingness Profile (MP) from complementary characteristics of the missing-data pattern, including missing rate, gap structure, temporal variability, sensor correlation, and operational context. It then derives a Missingness Complexity Score (MCS) that provides a compact decision signal for the selector. Based on MP and MCS, the framework routes the data to an appropriate imputation family, ranging from lightweight interpolation to machine-learning, sequence, cross-sensor, or generative methods. A quality-validation layer checks the reconstructed output and returns low-confidence cases for method re-selection. This creates a closed Select-Impute-Validate-Adapt loop rather than a fixed imputation pipeline. An environmental sensor case study provides initial evidence for the motivation of the framework and shows that common imputation methods do not remain equally reliable as missingness becomes more difficult. MIRA-IOT therefore contributes a decision and orchestration layer around existing imputation algorithms, with particular relevance to IoT, edge analytics, and Digital Twin data pipelines.

کلیدواژه ها:

Internet of Things ، missing data ، data imputation ، sensor networks ، adaptive selection ، Digital Twin