A tool for the survivability prediction of patients with colon or rectal cancer.

A tool, using Machine Learning techniques, for the survivability prediction of patients with colon or rectal cancer, up to five years after diagnosis and treatment and after some time after treatment. Please note that this app has not been tested in real life yet.

This work involved the use of different meta-classification schemes to construct survival prediction models, for colon and rectal cancer patients after treatment and after some time after treatment.

Model uses a Stacking classification scheme, combining k-NN, Decision Tree, and Random Forest classifiers as base learners and a Naive Bayes classifier as a stacking model learner. These models were trained with an unbalanced dataset, without unknown cases for the selected features.

To ensure that the model is able to adapt and adjust, an online learning server was created. In this way, it will be possible for users to dynamically feed new cases to the prediction system and make it change in order to provide better survival predictions.

Future work includes the continuous improving of the models and of the tool. The in- crease of the time windows, for the conditional survival calculators, is one of the proposals. In order to better understand comprehend the capacities of the developed tool, it will be interesting carrying out prospective tests.