ABSTRACT
In this chapter the authors successfully demonstrate the decoding and reconstructing of the common machine learning classification methods, such as linear discriminant analysis and k-nearest neighbor, from the underlying mathematics behind these algorithms, applied to two agricultural applications (soybean aphids and weed species) using open source platforms like ImageJ and R through developed user-coded programs. The decoded ReliefF algorithm in each dataset (soybean aphids, weed species, and iris, for comparison) selected the most influential features, among 13 shape-based features, for developing the reconstructed classification algorithms. Features such as roundness, solidity, and circularity for soybean aphids; diameter range, Feret aspect ratio, and solidity for weed species; and sepal length, petal length, and petal width for irises emerged as the best, based on weights.
