Hyperparameters and centroid improvements in the K-medoids method for grouping processed beef SMEs

Abstract

Hyperparameter tuning is a crucial step in finding the optimal machine learning parameters by iterative processing depending on the proper tuning, resulting in maximum accuracy. K-Medoids is a clustering algorithm susceptible to centroids and determining the correct K value to produce the low error. The most straightforward hyper-parameter technique used is Grid Search (GS) because it searches for deals using a specific range of distances. The K-Medoids method is used to form low-bias cluster models and make decisions on cluster development strategies that are appropriate to conditions in SMEs. Data from survey results is processed through the Label Encoding stages to convert categorical data into numerical data, Data Imputation to fill in empty values in the data using the mean method, and Data Normalization using the Min-Max method to standardize data in the range 0 to 1 for optimal processing. The results of cluster optimization using the GS method using a specific value range value is at the optimal number of clusters k = 10, with the lowest SSE value of 32,0970. When an increase in optimization for centroids in K-Medoids using a binary search algorithm results in an optimal cluster at k = 3 of DBI 0.1021 value. A comparison of the performance of the K-Means method and K-Medoids method shows that k-medoids produce the lowest SSE in the best parameter optimization through GS to improve model efficiency.