数据集获取

# 克隆仓库（含数据与代码）
git clone https://github.com/Ryan-McKee2001/Deep_Learning_Project-Breast-Cancer-Detection-NN.git

# 或仅下载打包文件
curl -L -o repo.zip https://github.com/Ryan-McKee2001/Deep_Learning_Project-Breast-Cancer-Detection-NN/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

Breast Cancer Diagnosis using Neural Networks

Project Overview

This project involves constructing and optimizing an artificial neural network to diagnose breast cancer using the Wisconsin dataset. The initial model is evaluated, and enhancements are proposed and implemented to improve classification accuracy and overall model performance.

Dataset

The Wisconsin Breast Cancer dataset contains 569 observations with 30 features each. It is used to train and evaluate the neural network model.

Model Architecture

• Input Layer: 30 nodes
• Hidden Layer: Initially 5 neurons, later increased to 15 neurons
• Output Layer: 1 neuron for binary classification
• Activation Functions: Initially sigmoid, later changed to ReLU
• Preprocessing: Standard normalization applied

Performance Evaluation

• 5-Fold Cross-Validation: Used to rotate data subsets for training and validation.
• Confusion Matrices: Used to evaluate precision, recall, F1 score, and accuracy.
• ROC Curve: Used to assess the model’'s discriminatory ability.

Baseline Model

• Training Accuracy: 0.728212
• Testing Accuracy: 0.732619
• Training Error: 0.568751
• Testing Error: 0.554755
• AUC: 0.5 (indicating poor performance)

Model Improvements

• Preprocessing: Applied standard scaling.
• Activation Functions: Changed hidden layer activation from sigmoid to ReLU.
• Hyperparameters: Adjusted batch size from 128 to 100 and learning rate from 0.01 to 0.03.
• Model Complexity: Increased the number of neurons in the hidden layer from 5 to 15.

Final Model

• Training Accuracy: 0.988577
• Testing Accuracy: 0.973622
• Training Error: Improved significantly
• Testing Error: Improved significantly
• AUC: 1 (indicating excellent performance)

Results

The final model exhibits significant performance improvements across all key metrics, providing a robust classification model for the Wisconsin Breast Cancer dataset.

• Precision: 0.977452
• Recall: 0.980992
• F1 Score: 0.978994
• Accuracy: 0.973622
• AUC: 1

Conclusion

Through systematic evaluation and enhancement, the neural network model for diagnosing breast cancer was significantly improved. The final model demonstrates excellent performance and provides a feasible solution for classifying the Wisconsin Breast Cancer dataset.

Acknow

数据说明

该数据集是一个用于乳腺癌诊断的公开医学数据集，包含569个样本观测值，每个样本由30个特征组成，这些特征来源于乳腺细胞核的数字化图像分析。数据集通过细针穿刺活检采集细胞样本，并经过专业病理学评估，主要用于训练和评估神经网络等机器学习模型在乳腺癌良恶性分类任务中的性能。其核心研究方向包括医学影像分析、疾病诊断辅助系统开发以及机器学习在临床医学中的应用。

完整仓库：github.com/Ryan-McKee2001/Deep_Learning_Project-Breast-Cancer-Detection-NN