Solving Multi-Class classification problems
Insurance agents are faced with the challenge of correctly identifying the Business Activity (BA) when selling. Major non-life insurers demanded attention to their dictionary of BA codes following the bundles of insurance policies that each would require. That presented a task – correctly identify the Business Activity (BA) code of a company when selling insurance to them.
The goal consisted of building a machine learning model to automatically classify the BA code of a company based on its text description. After several different approaches and types of algorithms tested, TFIDF (Term Frequency – Inverse Document Frequency) proved to be essential since it allowed to attribute weight to different features.
The solution resulted in 75% accuracy for the whole 6 digits of the BA code and considerably higher for the first 3 digits.