Classifying Learning Styles with Multinomial Naive Bayes

To explore whether a machine can determine how someone learns based purely on what they write, this project introduces a text classifier that predicts Visual, Auditory, and Kinesthetic (VAK) learning styles directly from behavioral sentences. Using a Kaggle dataset of over 15,000 English sentences, the pipeline converts text into numerical features via TF-IDF vectorization and processes it using a Multinomial Naive Bayes algorithm that was implemented entirely from scratch. Rather than outputting a single rigid label, the model calculates a nuanced probability breakdown for each learning style, achieving an impressive 87.57% overall accuracy. Ultimately, this demonstrates that transparent and relatively simple machine learning approaches can be highly effective without relying on complex deep learning, paving the way for future enhancements like bigram integration, dataset balancing, and an upcoming Indonesian-language version.