Optimized Data Mining Techniques for Outlier Detection, Removal, and Management Zone Delineation for Yield Prediction

Abstract

Enormous agricultural data collected using sensors for crop management decisions on spatial data with soil parameters like N, P, K, pH, and EC enhances crop growth for soil type. Spatial data play vital role in DSS, but inconsistent values leads to improper inferences. From EDA, few observations involve outliers that deviates crop management assessments. In spatial data context, outliers are the observations whose non-spatial attributes are distinct from other observations. Thus, treating an entire field as uniform area is trivial which influence the farmers to use expensive fertilizers. Iterative-R algorithm is applied for outlier detection to reduce the masking/swamping effects. Outlier-free data defines interpretable field patterns to satisfy statistical assumptions. For heterogeneous farms, the aim is to identify sub-fields and percentage of fertilizers. MZD achieved by interpolation technique predicts the unobserved values by comparing with its known neighbor-points. MZD suggests the farmers with better knowledge of soil fertility, field variability, and fertilizer applying rates.