Breast-Cancer-Detection

Problem Statement

Breast cancer is the most commonly diagnosed cancer and second-leading cancer death among women in the U.S. with 1 in 8 expected to develop it in her lifetime. Its overall death rate has steadily decreased by 1% per year from 2013 to 2018 which has thought to be from treatment advances and earlier detection.

Since earlier detection is thought to be a cause in decreasing breast cancer deaths, the predictive model built in this notebook serves as an attempt to fill this purpose as a classification problem. Preliminary analysis is done to make conclusions based on the data and select the best features for predictive accuracy.

Description

Pulled from the University of California at Irvine Machine Learning Repository’s Wisconsin Breast Cancer Database, cells were measured through a digitized image of a fine needle aspirate (FNA), a biopsy procedure where a sample is collected through inserting a needle. Because all samples contained more than one cell, their physical characteristics were averaged with respect to the amount of cells in the sample.

Diagnosis is a categorical variable defined in the data as:

• B - Benign (non-cancerous)
• M - Malignant (cancerous)

For classification purposes, the data will be formatted such that 0 represents benign and 1 represents malignant.

Thus, the goal of this project is to use machine learning classification methods to predict which samples are most likely to be benign or malignant.

Methods

• Machine Learning
• Data Analysis and Visualization
  • Unimodal
  • Multimodal
• Inferential Statistics
• Predictive Classification (Support Vector Machine)

Technologies/Tools Used

• Python
• NumPy, pandas
• scikit-Learn
• Jupyter
• Google Colab (not necessary, however if not used then dataset must be referenced locally with pd.read_csv())