Fuzzy Based Hybrid Imputation Of Missing Sensor Data For Reliable Smart Farming Analytics

Abstract

The accuracy and dependability of smart farming analytics are significantly compromised by missing sensor data, often caused by hardware malfunctions, energy constraints, or harsh environmental conditions. This paper introduces a novel fuzzy-driven hybrid imputation framework designed to enhance data completeness and semantic consistency in Internet of Things (IoT)-enabled agricultural monitoring systems. The core of the proposed method is the Agro-Fuzzy Adaptive Rule Engine (AFARE), which uses a Statistically Adaptive Semantic Partitioning (SASP) mechanism to construct fuzzy rules without requiring expert knowledge. These rules semantically infer missing values while respecting agricultural interpretability. To further improve robustness, the imputed outputs from AFARE are refined using a two-phase hybrid model: MissForest for global correlation modeling and K-Nearest Neighbors (KNN) for localized smoothing.Experiments were conducted on the Crop Recommendation V2 dataset with artificially induced missingness at 10%, 20%, and 30% levels. The AFARE-HIM model outperformed both traditional and advanced baselines. At 30% missing data, it achieved an RMSE of 0.411, R2 of 91%, and MAE of 0.82, surpassing MissForest (RMSE: 2.31) and SoftImpute (NIA: 86.9%).Visual comparison using scatterplots and error histograms further validated the structural alignment of imputed values with the true distribution.