Starting as a participation in BigData Competition

22.09 ~ 22.12

BigData Competition in NH Investment & Securities

• Started and led a project in BigData Competition in NH Investment & Securities
• Topic: Advanced Customer Profiling and Personalized Investment Portfolio Curation
• Devised a clustering technique to profile financial investment proclivity of internal 7348 customers data in NH Investment & Securities
  • Autoencoder: Linear, Convolutional
  • Dimensionality Reduction: t-SNE
  • Clustering: K-means
• Utilized high dimensional data, including features of risk-preference, income level, and financial securities preference, along with external stock price data from S&P Capital IQ and investpy package to incorporate volatility data based on customer stock balance information.
• Optimal Clusters, important features extracted by SHapley Additive exPlanations, and sample applications for financial service product are shown below.

Extended previous project into an academic research

23.01~23.08

Summary

• We present the high-dimensional data and feature set using novel network-based visualization methods and identify the multi-stage process’s optimal configuration.
• The approach segments 14,837 potential customers, each with 163 categorical and 143 numerical features from National Survey of Tax and Benefit Data from Korea Institue of Public Finance
• The first stage of the dimension reduction process employs deep neural network-based autoencoders.
• The second and third stage uses a non-neural network-based dimension reduction algorithm and clustering algorithm contingent on clustering performance.
• Subsequently, game theory-inspired Shapley values are computed for each feature to enhance explainability.
• The optimal approach involves an autoencoder, isometric mapping to three dimensions, and K-means clustering.

My Contribution