Analysis of Invoice Management System using Regression techniques with improved loss functions
Rakesh Kumar, Meenu Gupta, Pankaj Goplani, Abhijit · 2023
Regression models are a key tool in machine learning for predicting numerical values based on input features. In order to train a regression model, it is necessary to define a loss function that measures the difference between the predicted and actual values. The choice of the loss function has a defining effect on the model's performance, and different loss functions may be more appropriate for different types of problems. This paper provides an overview of the loss functions most commonly used for regression models, including mean squared error, mean absolute error, and Huber loss. The research also explores the mathematical properties of these loss functions, discusses their strengths and weaknesses in different contexts, and reviews recent research on new loss functions and their applications in specific domains. The analysis shows that the choice of loss function should be based on the data's characteristics and the model's specific goals. To analyze the model performance, 1,014,269 invoices have been collected, out of which 862,128 invoices were used for training, and 152,141 for validation. In the result analysis, tested models provided better accuracy with improved loss functions as opposed to conventional loss functions.