We developed a classification algorithm to support our client’s mission of educating people about investments using innovative technology. Our AI onboarding solution classifies users to determine their investor type. The process involved several steps: generating inputs, data labeling, algorithm decision-making, training, and testing.
We began by preparing a dataset of 1,000 surveys. The aggregated survey data was structured and preprocessed as a preparatory step for developing the risk tolerance portfolio classifier. Through label and one-hot encoding, various survey questions were transformed into numerical variables, with special attention to the implicit order of different answer categories.
The one-hot encoding approach increased the dimensionality of the training dataset, raising concerns about the adequacy of the collected samples. However, exploratory data analysis indicated that no additional samples were needed. The data exhibited near-linear separability and almost 100% accuracy, allowing for successful classification of low-, medium-, and high-risk tolerance portfolios based on survey responses.
Check here full case study: https://brigit.dev/case-studies/economia-ai