ML a Data-driven algoritmus pro personalizované řazení investičních příležitostí na základě chování investorů fintech platformy

This MBA thesis focuses on developing a ML algorithm for personalized ranking of investment opportunities on an investment platform. Specifically, it calculates the probability of investment for each investor across all available investment opportunities based on their historical investment behavior. The primary objective is to accelerate successful investment opportunity funding and identify key predictors of investment behavior of the investors on the platform. The thesis implements a predictive ML model using the XGBoost library in Python, which analyzes over 7.5 million historical user-project combinations. The model processes 36 features including investor demographic characteristics, investment opportunity parameters and behavioral data. The data infrastructure leverages the cloud-based Snowflake platform and Keboola for data preparation, and Python for implementation and visualization of the machine learning algorithm and its entire pipeline. The final model achieves an AUC score of 0.8902 and a lift factor of 2.91, meaning that there is 2.91x more investors in the group of predicted investors compared to random selection. Feature importance analysis identifies high user AUM, small project size, and high interest rates as the most significant predictors for investment of the user on the platform. The thesis encompasses a complete implementation from data preparation and exploratory analysis, through feature engineering and hyperparameter tuning to final model training and creation of a web application for results visualization. The business case anticipates an increase in investment conversion rate of 2-3 percentage points and company AUM growth of 15-20 %. The concluding section discusses regulatory challenges associated with algorithm implementation and offers recommendations for pilot deployment and a gradual rollout plan. The implemented solution provides a technological competitive advantage, as similar personalization algorithms are not commonly utilized in the alternative investment market.