Hi everyone, I’m working on improving Predictive Intelligence for incident routing (predicting Assignment Group) and would appreciate some advice.
Right now, I’ve built a classification solution using NLP on the Short Description and Description field. I’ve applied filters to improve data quality, which reduced my dataset from about 127K records to 22K high-quality records, and I’m currently getting around 83% precision.
I’m looking for guidance on how people typically improve prediction results beyond this point. What tactics or best practices have worked for you? For example, are you seeing better results by adding more input fields, adjusting filtering strategies, tuning training data, or using multiple models?
Any practical tips, approaches, or lessons learned would be really helpful.
