A Novel Catboost-Categorical Cross Square Entropy Method for Binary and Multiclass Classification

Abstract

The Organization of data based on the types of information it contains is challenging, and it is critical to manage unstructured textual data from varied domain. In addition to many classification algorithms like Support Vector Machines, Random Forests, Decision Trees, and Logistic Regression, a large number of them employ boosting techniques like Adaboost, XGboost, and LightGBM. Despite classifying the data using the catboost method, we are aiming to obtain 100% accuracy by further refining the model. So here we propose a novel algorithm called Catboost Categorical Cross-Square Entropy (CCCSE) to improve performance. Using the catboost categorical cross square entropy, the loss function, which is ultimately for accurate data classification, is constructed. A categorization procedure is employed to arrange the dataset’s 12, 94, 772 records. The experiment is carried out for both classifications using a variety of datasets obtained from Kaggle, Github, Google Scholar, and the IEEE data. It is preferable to use the proposed CCCSE method for the loss function rather than the categorical cross-entropy for multiclassification, and sigmoid cross entropy for binary classification respectively.The proposed CCCSE formula for binary and multiclass classification classifies document with 100% accuracy than the existing state of the art methods.