Breast Cancer Prediction

01 · Purpose

Using patterns to support a difficult decision

I approached this project as an introduction to machine learning in a high-stakes setting. The goal was to classify breast cancer records as malignant or benign using structured diagnostic features.

What interested me was not the idea of replacing clinical judgment. It was understanding how a model could organize many measurements into a consistent signal that might support further review.

02 · Preparing the data

Making the dataset understandable first

The work began with data cleaning, removing empty fields, encoding diagnosis labels, and studying feature distributions and correlations. I then scaled the features and separated the data into training and testing sets.

This stage reinforced a simple lesson: a model cannot rescue an unclear or poorly prepared dataset.

03 · Comparing models

Learning from more than one answer

I compared Logistic Regression, Decision Tree, and Random Forest models rather than treating the first result as final. Each model was evaluated with confusion matrices, classification reports, and accuracy.

The comparison helped me understand the trade-off between a simpler interpretable baseline and more flexible tree-based methods.

04 · Reflection

Accuracy is not the whole story

In a medical classification problem, the consequences of different errors are not equal. A false negative can matter far more than a false positive. Looking back, I would make recall, class-specific errors, calibration, and interpretability more central to the evaluation.

The project became an early point of interest in how machine learning can support decisions without hiding uncertainty.

05 · The next question

Toward interpretable decision support

A future version could combine structured measurements with imaging, pathology text, or patient history, provided the data and governance were appropriate.

I would also add explainability, uncertainty estimates, and clinician-centered evaluation. Healthcare is an area I am interested in exploring further, but any future work would need to stay grounded in clinical evidence and responsible use.