Comparative Analysis of Machine Learning Algorithms and Statistical Techniques for Data Analysis in Crop Growth Monitoring with NDVI

Abstract. We assessed the potential of Machine Learning (ML) for mapping crop growth in three flood irrigated fields. Results generated from ML algorithms were compared to the output generated by the ISODATA algorithm. Affinity Propagation (AP) identifies the number of clusters by considering all data points as potential exemplars and iteratively refine the set, while Gaussian Mixture Model (GMM) algorithm treats the data as a mixture of several Gaussian distributions, allowing for flexible cluster shapes. In contrast, ISODATA, a statistical clustering method, requires an analyst to specify the number of output clusters followed by iterative splitting and merging of clusters based on variance and distance criteria. We acquired Landsat derived NDVI images for three flood-irrigated fields over a span of four years. These images were collected at the start of the growing season to ensure consistency. Initially we clustered the pixels in these images for each field using AP and determine the number of clusters. Next, we applied GMM to identify and define the clusters. Finally, we plotted the mean value of all the pixels in each cluster for every year and assigned the clusters into six thematic classes: the first three classes for consistent growth (good, average, or poor) across all four years, and the other three for mixed growth patterns (e.g., good in three years and average in one). Output maps generated from these methods were compared using IoU scores. ML methods had greater efficiency in terms of replicating the steps for other fields, whereas ISODATA requires analyst intervention and interpretation.