文件列表:

part1.ipynb
part1.py
part2.ipynb
part2.py

# K-Nearest Neighbor Algorithm Implementation ## Part 1: Classification of News Articles ### Dataset The dataset for Part 1 can be found [here](https://www.kaggle.com/datasets/qusaybtoush1990/english-news), containing 1491 samples with 5 discrete ground-truth class types: Sport, Business, Politics, Entertainment, Tech. ### Approach #### 1.1 Understanding the Data Implemented a nearest neighbor algorithm to predict the category of news articles using a Bag of Words (BoW) model. Feasible keywords, such as 'election,' 'technology,' and 'profit,' were analyzed for their frequency in each category. #### 1.2 Implementing k Nearest Neighbor - Utilized the k Nearest Neighbor algorithm to classify articles. - Represented data using BoW with options for Unigram and Bigram. - Handled unseen words during testing by either ignoring or assigning a non-zero default value. #### 1.3 Analyzing the Words 1. Listed 10 words predicting the article's category presence and absence for each category. 2. Explored TF-IDF and Information Theory for word selection. #### 1.4 Stopwords Identified 10 non-stopwords predicting article categories for each category and discussed the relevance of removing or keeping stopwords. ### Classification Performance Metric Computed Accuracy, Precision, and Recall for each test using 5-fold cross-validation. Reported average metrics. ### Error Analysis for Classification - Identified misclassified samples and discussed challenges. - Compared feature choices, system parameters (e.g., k), and computation time. ## Part 2: Medical Insurance Cost Estimation from Data ### Regression Dataset The dataset for Part 2 can be found [here](https://www.kaggle.com/datasets/mirichoi0218/insurance), with 1338 samples and continuous medical cost values. ### Regression Performance Metric Calculated Mean Absolute Error (MAE) for each test using 5-fold cross-validation. Reported average MAE. ### Feature Normalization Used min-max normalization to re-scale feature columns between 0-1. ### Error Analysis for Regression - Compared feature normalization choices. - Investigated the effect of system parameters on performance. - Discussed computation time in addition to regression rates. ## Implementation Details - Implemented K-fold cross-validation, shuffle methods, k-NN, weighted k-NN, accuracy, precision, recall, MAE, and min-max feature normalization without using ready-made libraries. - Used Numpy array functions for intermediate steps.

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：