AI & Product · 2022
NLP for automated IT ticket routing — 91% classification accuracy
AI-generated summary of a LinkedIn post by Pranav Kumar Singh — not a verbatim reproduction. Read the original on LinkedIn.
The post describes a project addressing inefficient manual ticket allocation in IT support: roughly 8,500 support tickets were being incorrectly assigned to the wrong functional group about 25% of the time, causing avoidable delays in resolution.
The methodology moved through several stages — analyzing how tickets were distributed across functional groups, addressing class imbalance in that distribution through resampling, cleaning ticket descriptions and translating non-English content, then standard NLP preprocessing (stop-word removal, lemmatization) before classification.
Both supervised ML classifiers (SVM, KNN, XGBoost) and neural approaches (LSTM, transfer learning) were tried. The finding worth highlighting: the supervised ML classifiers clearly outperformed the neural network models — 77% and 91% accuracy versus roughly 65% for the simple neural approaches — a useful data point that more complex deep learning isn't automatically the better choice for structured text classification at this scale.
The project was a collaboration with Bhawna Jha and Janki Pandya, mentored by Dipanshu Haldar, with the GitHub repository linked in the post's comments.