Breast Cancer Prediction Model

Table of Contents

• Description
• Source
• Dataset Details
• Installation Guide
• Model and Methodologies
• Outcomes and Evaluation
• Improvements
  • Tune KNN Model Using RandomizedSearchCV for Hyperparameter Optimization
  • Implement MICE for missing value imputation
  • RobustScaler to Scaling Methods
  • Implement Lasso Regression for feature selection
• How to Use
• Future Directions

Description

The Breast Cancer Wisconsin (Original) Dataset is a well-known dataset used for binary classification tasks to predict the diagnosis of breast cancer. It contains cases derived from fine needle aspirate samples, providing critical information to help differentiate between benign and malignant tumors.

Common Use Cases

• Predicting whether a breast mass is benign or malignant using machine learning algorithms.
• Identifying the most relevant features for classification.
• Evaluating the performance of different machine learning models on a standardized dataset.

Limitations

• The dataset's simplicity may not reflect the complexity of real-world scenarios, limiting generalization.

Source

You can download the dataset from the UCI Machine Learning Repository: Breast Cancer Wisconsin (Original) Dataset

Dataset Details

• Sample Size: 699 samples
• Features:
  • ID number: An identifier for each sample.
  • Clump Thickness: A measurement of the thickness of the clump of cells.
  • Uniformity of Cell Size: Measures how uniform the size of the cells is.
  • Uniformity of Cell Shape: Assesses how uniform the shape of the cells is.
  • Marginal Adhesion: The adhesion of the cells at the margins.
  • Single Epithelial Cell Size: The size of a single epithelial cell.
  • Bare Nuclei: The presence of nuclei without surrounding cytoplasm.
  • Bland Chromatin: The texture of the chromatin in the cell nucleus.
  • Normal Nucleoli: The presence of normal nucleoli in the cells.
  • Mitoses: The count of cells undergoing mitosis.
• Target Variable: Class label indicating whether the tumor is benign (2) or malignant (4).
• Data Format: CSV

Installation Guide

To run this project, ensure you have Python installed along with Jupyter Notebook. You'll also need the following libraries:

• pandas
• numpy
• scikit-learn
• matplotlib
• seaborn

To install these libraries, you can use pip:

pip install pandas numpy scikit-learn matplotlib seaborn 

Model and Methodologies

In this analysis, the K-Nearest Neighbors (KNN) algorithm is used for modeling and classification purposes. Key steps include:

• Data pre-processing: Handling missing values, detecting outliers, data normalization and feature selection to identify the most significant attributes, and splitting the data into training and testing sets.
• Model training: Using a K-Nearest Neighbors (KNN) classifier to train the model on the training set.
• Model evaluation: Assessing the model's performance using accuracy, precision, recall, F1-score, classification report, confusion matrix, and AUC score.

Outcomes and Evaluation

The model achieved an accuracy of 94% on the test set. Key performance metrics include:

• Precision: 0.95
• Recall: 0.93
• F1-score: 0.94
• AUC score: 0.94

Detailed performance metrics and visualizations are available in the results section of the repository.

Improvements

This section aims to showcase several more robust models and data preprocessing techniques developed in this project, designed to enhance both the accuracy and interpretability of the results.

1. Hyperparameter tuning for the K-Nearest Neighbor (KNN) algorithm.
2. Implement MICE for missing value imputation
3. RobustScaler to Scaling Methods
4. Implement Lasso Regression for feature selection

Tune KNN Model Using RandomizedSearchCV for Hyperparameter Optimization

In this updated version of the model, I have enhanced the K-Nearest Neighbors (KNN) algorithm through hyperparameter tuning using RandomizedSearchCV. This tuning method allows for a more efficient exploration of hyperparameter space, leading to the identification of optimal settings that improve model accuracy and performance.

Hyperparameter Tuning Details

• n_neighbors: Tested a range from 1 to 10 to determine the most effective number of neighbors for classification.
• weights: Evaluated both 'uniform' and 'distance' weighting to analyze their impact on predictions.
• metric: Explored multiple distance metrics, including 'euclidean', 'manhattan', and 'minkowski'.
• algorithm: Compared various algorithms such as 'auto', 'ball_tree', 'kd_tree', and 'brute' to find the most efficient method for neighbor search.
• p_value: Incorporated values for the Minkowski distance power parameter, specifically 1 for Manhattan and 2 for Euclidean distances.
• leaf_size: Adjusted the leaf size for tree-based algorithms, examining a range from 1 to 21.

Results Overview

Despite the comprehensive tuning efforts, the model's performance metrics remained consistent with previous evaluations:

• KNN Testing Accuracy: 94.12%
• KNN Tuning Accuracy: 94.12%
• AUC Score: 94.15%
• AUC after Tuning: 94.15%

These results indicate that, while tuning practices are beneficial for exploring the potential of the model, this particular process did not yield an improvement in accuracy. This emphasizes the robustness of the initial settings and highlights the ongoing challenge of optimizing model performance in machine learning. Future work may explore additional feature engineering or alternative algorithms to further enhance predictive capabilities.

Implement MICE for missing value imputation

A recent update includes a method to handle missing values using Multiple Imputation by Chained Equations (MICE). MICE allows for a more robust handling of missing data by generating multiple imputed datasets and combining the results for improved accuracy and reliability. This approach enhances the overall model performance and provides more credible predictions, making it particularly beneficial for analyses where missing data is a concern.

RobustScaler to Scaling Methods

In this update, I have incorporated the RobustScaler scaling method into my Breast Cancer Wisconsin machine learning model.

RobustScaler is a normalization technique that scales features using statistics that are robust to outliers, specifically the median and the interquartile range (IQR). This makes it particularly useful for datasets containing outliers, as it centers the data around the median and scales it according to the IQR, thereby reducing the influence of extreme values.

While the RobustScaler method is beneficial for improving the performance of models on datasets with outliers, the results from this implementation did not exceed the performance of prior scaling techniques. The results achieved are as follows:

• Z-Standard Scaling: 0.9529
• Min-Max Scaling: 0.9441
• RobustScaler Scaling: 0.9382

The RobustScaler result of 0.9382 indicates that, although it provides a robust solution against outliers, it did not improve the model's performance compared to Z-standard and Min-Max scaling methods.

Implement Lasso Regression for feature selection

This update applies Lasso Regression for feature selection on the Breast Cancer Wisconsin dataset. Lasso Regression (Least Absolute Shrinkage and Selection Operator) is a linear regression technique that utilizes L1 regularization. This method not only decreases the complexity of the model but also encourages sparsity by shrinking some coefficients to zero. This characteristic makes Lasso particularly suitable for feature selection, as it effectively selects a subset of features while reducing the risk of overfitting. Key benefits include:

1. Automatic Feature Selection: Retains only relevant features.
2. Interpretability: Simplifies the model for easier understanding.
3. Reduced Overfitting: Focuses on critical predictors.

Results

The Lasso Regression method achieved an accuracy of 93.21%, slightly lower than the previous method, Recursive Feature Elimination (93.44%). Lasso is effective for feature selection, enhancing model interpretability and avoiding overfitting, even if it didn’t outperform RFE in accuracy.

• Update Note:
  Also in this update, the ROC curve for the KNN model has also been refreshed to reflect the latest results and provide a clearer comparison of model performance.

How to Use

To use this project, clone the repository using the following command in your terminal or command prompt:

1. Clone the repository:

   git clone https://github.com/Ehsan-Behzadi/Breast-Cancer-Prediction-Model.git  
   cd Breast-Cancer-Prediction-Model

Next, open the Jupyter Notebook file (usually with a .ipynb extension) using Jupyter Notebook.

2. To start Jupyter Notebook:

   jupyter notebook

Future Directions

Future improvements and directions for this project include:

• Exploring other classification algorithms such as Random Forest, Support Vector Machine (SVM), and more.
• Hyperparameter tuning to optimize model performance.
• Incorporating additional features to enhance prediction accuracy.