MATLAB Implementation of Naive Bayes Classifier with Code Description

The following describes the MATLAB code implementation and dataset specifications: 1. The dataset consists of text data. Therefore, class labels are numerically encoded according to their top-down order in the table using digits 1, 2, 3...15. For example, the topmost class "diaporthe-stem-canker" is represented by digit 1. Attribute names are similarly encoded with sequential numbering starting from 0. For instance, when data=april, the corresponding value in the data table is encoded as 0. The MATLAB implementation includes preprocessing functions to handle this categorical-to-numerical conversion automatically. 2. The experimental program is strictly implemented according to the Naive Bayes principle. This probabilistic classification method operates under the assumption of feature independence. The MATLAB code implements the core Naive Bayes algorithm using probability density estimation functions, with separate modules for training (probability calculation) and prediction (classification). 3. The total sample size comprises 290 instances divided into 15 classes. The dataset is partitioned using a 75%-25% split strategy, where approximately 75% of samples per class are allocated for training and the remaining 25% for testing. This results in 218 training samples and 72 test samples. The code includes data partitioning functions that maintain class proportions during splitting. 4. All probabilities calculated according to the Naive Bayes principle are derived from frequencies in the training samples. For example: a) The probability of different attribute values within each class (illustrated with data=4 in the diaporthe-stem-canker class): p(data=4|diaporthe-stem-canker) = (frequency of data=4 in diaporthe-stem-canker) / (total samples in diaporthe-stem-canker) The implementation uses frequency counting functions and probability lookup tables for efficient computation during prediction. b) The probability of each class label (using diaporthe-stem-canker as example): p(diaporthe-stem-canker) = (samples in diaporthe-stem-canker) / (total training samples) The trained probability models are applied to classify test samples, achieving a final classification accuracy of 77.7778%. The prediction module implements the Bayes theorem calculation using logarithmic probabilities for numerical stability. Note: The Naive Bayes implementation principle used in this experiment is adapted from online blog resources, with custom MATLAB optimization for handling categorical data encoding and probability smoothing techniques.