Machine learning and statistical inference based data imputation technique for crop disease data

Crop disease detection is one of the precision agriculture tasks that predict crop types from large amounts of heterogeneous, dimensionally oriented data. Crop disease data is collected from sensors, weather stations, satellite imagery, UAVs, and expert observations. The accurate dataset improves the accuracy of crop disease detection. The missing value is one of the significant problems in the crop disease dataset that minimizes the accuracy rate of the crop disease detection. The missing values occur during data collection. The sensor failures, communication gaps, environmental disturbances, and manual recording errors are the more common causes of missing values. The inadequate handling of missing data leads to biased results in statistical analysis and unreliable implementation of disease prediction models. The imputation approach is one of the solutions for handling missing values that improves the statistical analysis and eases the prediction model. The missing data are categorized as completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). The mean, mode, median, and sampling are the more popular statistical-based data imputation methods. The logistic regression, support vector machine, K-nearest neighbor, naïve bayes, decision tree, random forest, gradient boosting machines, extreme gradient boosting, and artificial neural network are more usable machine learning-based data imputation techniques. This study analyses various statistical inference and machine learning-based data imputation techniques for crop disease datasets. The review identifies research gaps in crop disease dataset-based imputation frameworks to improve disease classification accuracy.