On predictive modeling of the twitter-based sales data using a new probabilistic model and machine learning methods

Abstract

The term marketing refers to the various strategies employed by a company to enhance the visibility of its brands among potential consumers. Advertising serves as an effective channel for marketing efforts, allowing a company to showcase its products and services to the target audience. Numerous probability-oriented strategies have been suggested and carried out to investigate the effectiveness of advertising mechanisms. In accordance with this pivotal portion of the literature, we develop a new probabilistic model called a new cosine inverse Weibull (NCI-Weibull) distribution, which serves to analyze the sales related to advertising on the Twitter platform. The derivation of point estimators is accomplished for the new model. In addition, a simulation study is presented, which is based on the NCI-Weibull distribution. We further illustrate the applicability of the NCI-Weibull distribution through an analysis of Twitter-based sales data and comparing it with various other statistical models. Furthermore, we employ two machine learning techniques to forecast the sales, with a particular emphasis on 1-step and 3-step advance predictions. The statistical analysis indicates that multilayer perceptron (MLP) is superior in the field of short-term forecasting, whereas support vector regression (SVR) is more effective in the context of longer-term predictions.

Keywords

• Inverse weibull distribution
• Cosine function
• Monte carlo simulation
• Twitter
• Sale
• Statistical modeling