Methodology of potential insurance buyers classification in networks of physical and online retailers

• Grant number: 18/15321-0
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: August 01, 2019
• End date: April 30, 2020
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Renato José Ferreira
• Grantee: Renato José Ferreira
• Company: Sutservices Soluções em Tecnologia Eireli
• CNAE: Outras atividades de serviços financeiros não especificadas anteriormente
• City: Barueri
• Pesquisadores principais: Eduardo Mayer Fagundes
• Associated scholarship(s): 19/07493-9 - Methodology of potential insurance buyers classification in networks of physical and online retailers, BP.TT; 19/08670-1 - Methodology of potential insurance buyers classification in networks of physical and online retailers, BP.TT; 19/09695-8 - Methodology of potential insurance buyers classification in networks of physical and online retailers, BP.TT

Abstract

The objective of this experimental research is to make a technical-scientific feasibility analysis of the use of Bayesian Networks to extract relevant information from consumers of mass products to train Deep Learning systems for predictive analyzes in the insurance market. The challenge is from sparse retailer information, associated with the need to preserve sensitive customer information, extract relevant data to train artificial neural networks to classify and predict the likelihood of a consumer acquiring insurance. Without this, artificial intelligence systems are inefficient. The problem is how to identify relevant data, considering that only 1% of products are purchased with insurance, making most traditional statistical methods inadequate. Researchers Neill, Moore and Cooper (2005) faced the same problem to detect outbreaks of emerging diseases. They solved the problem by developing a method, entitled "Bayesian Spatial Scan Statistics", based on the Bayesian method. The hypothesis of this project is that this method can extract the relevant insurance sales data from retailers to be used in training an artificial neural network to classify and predict the purchase of consumer insurance. It is important to monitor store insurance sales increases as early as possible to identify which factors influenced consumers and quickly match these factors in other stores to consumer groups with the same characteristics. Also, use these successful events to train the learning systems. The overall objective is to broaden the insurance market by using new information technologies to make it easier to take out policies and protect consumers' assets. According to Tarciso Hüber (2016), there is opportunity for up to 7% increase in retailers' sales of extended warranty insurance sales. The proposed methodology consists of four stages: 1. Step 1: Identify consumer clusters that purchased product insurance using the Bayesian method for detection of spatial clusters, developed by Neill, Moore and Cooper (2005); 2. Step 2: Analyze the profile of consumers and products of space clusters using quantitative and qualitative methods; 3. Step 3: Development of the artificial neural network using the Multilayer Perception with Backpropagation method to classify the consumers of spatial clusters with the highest probability of buying insurance; 4. Step 4: Create and adapt insurance offerings to the classified consumer profile and monitor the process with the PDCA continuous improvement method. (AU)