DRMF: Optimizing machine learning accuracy in IoT crop recommendation with domain rules and MissForest imputation

Abstract

In the realm of IoT-driven precision agriculture, addressing missing data is crucial for reliable crop recommendation systems. This paper proposes the Domain Rules and MissForest (DRMF) algorithm to handle the above mentioned challenge. The proposed DRMF algorithm was thoroughly tested on an IoT agriculture dataset with the introduction of a missingness mechanism in the form of MAR with 10 % of missing values. A comparison analysis with the usual imputation techniques such as Mean Imputation, kNN Imputation, Linear Regression, EM Algorithm, Multiple Imputation, and the standard MissForest was performed and the proposed method was found to perform better. The DRMF algorithm attained an unmatched Root Mean Squared Error (RMSE) value of 0.025 and a Mean Absolute Error (MAE) value of 0.012, displaying a significant superiority over its competitors. It is important to note that the algorithm also achieved a Mean Absolute Percentage Error (MAPE) of 5.0% and an R-squared value of 0.970, with the overall accuracy rate being 99.0%. The quantitative findings serve to emphasize the effectiveness of the DRMF algorithm in improving the prediction accuracy of crop recommendation models. The novelty of this research is in the combined approach that merges the computational power of the MissForest algorithm, and the insight offered by domain-specific agricultural rules.