An Anomaly Detection Algorithm for Estimating Interannual Variation in Crop Yields

Crop production is a vital aspect of human survival. Reducing global poverty heavily relies on increasing the robustness in crop yields. The quantity of crops harvested from a given agricultural land, also known as crop yields of that land is the important parameter that helps to increase the smartness in adopting novel agricultural plans. Measuring this parameter can be done through supervised machine learning techniques which helps with accurate predictions given the historical data of crop yield. This paper proposes an ideation, that uses Support Vector Machines (SVMs) because of its strength being accuracy, robust and flexible. SVM can predict reliable crop yield estimates that help to allocate resources and smart planning for agricultural activities. Also, SVMs are good enough to handle the imperfections in historical crop yield datasets, outliers, and inconsistencies. Furthermore, SVMs can bring out the relationships and any nonlinear patterns in the crop yields which are influenced by uncertain environmental factors over various seasons. This helps to identify the anomalies in the historical data. The anomalies could be exceptionally high yield or the failures which happen to be pinpointed as abnormal years. The identified outliers are excluded by the SVM, thus bringing out a prediction model that focusses on “normal years”. By this, it is possible to represent more accurate inter-annual trends of crop yields avoiding the anomalies aiding the farmer to make better decisions for much more productivity in the crop yield. Based on the ideation, simulations are performed that showcase the success of the SVM in predicting good for the crop yields in inter-annual period.