Combining Instance Weighting and Fine Tuning for Training Naïve Bayesian Classifiers with Scant Training Data

This work addresses the problem of having to train a Naïve Bayesian classifier using limited data. It first presents an improved instance-weighting algorithm that is accurate and robust to noise and then it shows how to combine it with a fine tuning algorithm to achieve even better classification accuracy. Our empirical work using 49 benchmark data sets shows that the improved instance-weighting method outperforms the original algorithm on both noisy and noise-free data sets. Another set of empirical results indicates that combining the instance-weighting algorithm with the fine tuning algorithm gives better classification accuracy than using either one of them alone.