UBCpredict

This python repository aims to develop an efficient urinary bladder cancer prediction model with machine learning techniques.

Simple steps to get it to work!

• Clone the repository
• Run "py predict.py"
• Select the model (from 1-5)
• Select filename for test dataset
• See the magic!

Prediction (predict.py)

This script is used to predict a model created by (create_model.py). Some models are already created by the work done in the paper referenced at the bottom. Create a csv file have 12 features per column (see example in test_dataset.csv). The features are explained below:-

• Age recode (value from 1-18: 1 = 1-4 years old, 2 = 5-9 years old,.., 18 = 85+ years old)
• Sex (1 = male: 0 = female)
• Year of diagnosis (actuall year, for years greater than 2015, just use 2015 instead)
• Martial status (1 = not married, 0 = married,seperated,divorced)
• Cancer grade (1-17)
• Tumor size (0-99mm = 100, 100-199mm = 200, ..)
• Lymph nodes (1-6)
• Total number of insitu tumors (1,0)
• Histologic type ICD-O-3
• Primary site code (lookup SEER documentation)
• Derived AJCC
• Regional positive nodes

Creating own model (create_model.py)

This script could be used to create your own models and contriute towards this repository.

Classifier comparison (algo_comparision.py)

As described in the paper, 4 algorithms were compared out of which best one was used as a default model in the (predict.py) file. This script runs a comparision test and prints out the metrics for evaluation of various classifiers.

Reference paper

The following paper was used in creation of this repository

/papers/prediction_of_the_morality_for_ubc.pdf

License

MIT License