While core collections offer various advantages, achieving a balance between representing genetic diversity and ensuring the practical manageability of the entire genebank collection is crucial. This study investigated the utility of the Shannon-Weaver diversity index and machine learning as a support system for the acceptability of new maize accessions into the genebank collection. This study examined 1279 maize germplasm accessions from the Agricultural Genebank Indonesia. The maize germplasm collection was divided into two parts. The first part, namely subset A, contains 600 accessions, which acted as original collections and were randomly selected for calculating the diversity values in each kernel character using the Shannon-Weaver diversity index. The second part, subset B, consisting of 679 accessions, was used to determine whether each accession was similar or different from subset A using an Excel macro-based application built by the authors. Principal component analysis (PCA) and Mahalanobis distances were used in the first step to identify outliers in data points with nine independent variables, namely kernel type, kernel color, presence of white cap, mottled type, kernel upper surface shape, kernel weight, kernel length, width, and thickness. Seventeen outlier samples, detected through PCA and Mahalanobis distances, were intentionally excluded from the dataset to ensure the integrity of subsequent machine learning analyses. The subset B was then divided into two parts to perform conventional and novel machine learning, i.e., linear discriminant analysis (LDA) and Tabnet. The results of this study show that the model's accuracy of LDA and TabNet were 89.36% and 86.4%, respectively. By integrating the Shannon-Weaver diversity index and machine learning methodologies, this research offers a comprehensive decision support system for guiding the acceptance of new maize accessions. The proposed system implies optimizing genebank strategies for the integrity and adaptability of maize germplasm collections.