The project is developing unsupervised machine-learning models to group customers into segments for the purpose of giving insurance product recommendations. Customers are divided into subgroups based on some types of similar characteristics. The dataset includes summary information on 18 behavioural variables from the 8,950 active credit cardholders. Behaviours include how a customer spends and pays over time. The notebook explores different unsupervised algorithms such as k-means, hierarchical clustering, and DBSCAN for an insurance company to divide customers into groups to optimize marketing campaigns for insurance products. Standardization is used to rescale data to have a mean of 0 and a standard deviation of 1. PCA and TSNE methods are used for dimensionality reduction and visualization. After comparing with the silhouette score and visualized plots, the optimal model is the k-means method with a k value of three that is trained with PCA scaled data. There are small groups of people who have similar behaviours on purchasing, cash advances, credit limits and so on. The K-means clustering method helps identify the group that has similar features. After the segmentation, an insurance company will provide insurance product recommendations based on their characteristics.
