top of page

← Back to all projects

Completed
Wellesley High School

Breast Cancer Predication Algorithm with Java

Using AP CSA (Computer Science Advanced) knowledge to create an algorithm that can categorize breast cancer tumors as low, moderate, or high risk of malignancy.

TIMELINE
2026
FIELD
Data science & 
Computer science
ROLE
Student Researcher
STATUS
Completed
PROJECT OVERVIEW

In this project, I sourced a famous breast cancer dataset from the University of Wisconsin to help classify a new tumor as low, medium, or high risk based off the following variables:
1.  Area
2. Texture
3. Concavity
4. Concavity points

Though I also considered:
5. Smoothness
6. Perimeter
7. Radius

Click HERE to find the dataset I used on Kaggle.

In the dataset, most variables are in decimal form (ie. smoothness = local variation in radius length, concavity = severity of concave portions of contour), which makes them numerically interpretable. Input values correspond to features extracted from digitized images of fine needle aspirates (FNA) of breast masses in the Wisconsin Diagnostic Breast Cancer dataset.

My program has two main classes: CancerAnalysis and Patient. The prior contains the main class and performs all the computations, while the latter creates an object for each patient with their individual parameters. 

CODE OVERVIEW

Though I'd love to ramble about everything that went into this project, it contains many methods and hundreds of lines of code, so I will summarize general ideas. 

For any new data analysis, the first and perhaps most important step is filtration. However, this dataset provided no null values or extremities, so I was free to use the data as delivered.

To begin, I looped through the entire dataset, creating an object for each patient. I used Java's nextLine() and split() functions to assign each relevant variable (ie. tumor radius) to a new parameter within the object. All of the patient objects were appended to an ArrayList called "patientlist," called later when appropriate.

For a user running the program, I offered two main options: view dataset metrics, or classify a new tumor.

The first option calculates the mean of each variable for malignant vs. benign tumor cells and offers some insight behind the scoring system. 

The second option uses the dataset to make a low/medium/high risk prediction for a new tumor. I embedded a part of the algorithm here––try it out!

THE SCORING SYSTEM

As specified in the program, the scoring method used in this program is moderately capable of marking benign tumors as low risk and highly capable of marking malignant tumors as high risk.

The scoring system could be shifted slightly to increase overall accuracy, but is meaningfully set up this way in order to ensure that malignant tumors do not go understated.

Approximately 42.2% of truly benign tumors in this dataset are labeled low risk and about 5% are labeled high risk, which results in some false alarms. However, 90.6% of truly malignant tumors are correctly labeled as high risk, with the rest classifying as medium risk.

This is especially important for  clinical usage, as it is far better to have false positives than to risk not treating a malignant tumor properly.

HOW ARE TUMORS SCORED?

Before arriving at the final method, I tried out various scoring options to see which wielded the best results.

I concluded that radius and perimeter were redundant with area, and that smoothness was not a significant factor, so I did not consider those variables when scoring.
 

The scoring system goes from 0-7, with higher scores corresponding to higher risk. The scale is:
0 - 1 = low risk
2 - 4 = medium risk
5 - 7 = high risk

Variables earn points depending on where they stand compared to low/high cutoffs. These cutoffs are determined by averages––the low cutoff is the average for the benign tumors in the dataset, while the high cutoff is the average for the malignant ones. However, the variables are weighed a bit differently, so they are not scored the same in relation to the cutoffs. The table below shows exact scoring:

Screenshot 2026-08-11 at 12.12.15 AM.png
FUTURE WORK

This program was submitted as my final project for AP CSA (Computer Science Advanced) at my high school, so I had limited time and resources to complete it. If I had more time, here are some improvements I could have made:
- Adjusted scoring to better represent low-risk tumors
- Learned how to integrate machine learning models, completely abandoning the averaging process (though this wouldn't be within the scope of AP CSA)
- Tested other variables within the dataset to see if they had a more significant transcriptional effect than initially assumed

Overall, however, I am proud of how the project turned out and the effort that went into it.


Thank you for reading! Contact me if you have any questions.
 

← Back to all projects

Drop Me a Line, Let Me Know What You Think

© 2035 by Train of Thoughts. Powered and secured by Wix

bottom of page