Tumor Classification using Machine Learning Classifiers

Queensland University of Technology Project: CAB320 Artificial Intelligence, completed in Semester 1 of 2020 (Feb' - June).

Summary

The aim of this project was to build four different types of classifiers and present their performance in a report. The classification task was to predict whether a tumour would be malignant or benign, denoted by M and B, respectively.

Types of classifiers and pre-selected hyperparameter;

1. Nearest Neighbour: number of neighbours,
2. Decision Tree: maximum depth of the tree,
3. Support Vector Machine: parameter C,
4. Neural Network: number of neurons in the hidden layer.

The maths behind each classification algorithm was not the subject of this work, the implementation of each classifier using Scikit-learn libraries was.

The works was separated into two steps;

1. Pre-processing the data. Prepossessed data set to remove any anomalies and convert to NumPy arrays. Separate patient diagnosis and patient test results into response and feature variables, respectively.

2. Building the classifiers. Each classifier follows the below pseudo-code with the appropriate libraries and functions used.

• Initialise the classifier from the sklearn library,
• Execute the grid search for hyperparameter optimization,
• Fit the data to the estimator,
• Re-train the estimator using the optimized parameter,
• Re-fit the data to the estimator.

By Harry Akeroyd. [email protected]