Multi-Class SVMs for Classifying Thyroid Diseases

Implemented a Support Vector Machine model for diagnosing thyroid disease.

• Specifically, the SVM is used to distinguish between 4 different thyroid states:
  • Hyperthyroid
  • Hypothyroid
  • Euthyroid-sick
  • Euthyroid (negative)

Motivation

• In the U.S. alone, an estimated 27 million Americans have Thyroid disease
• More than half of this population are undiagnosed or misdiagnosed
• Women are seven times more likely to contract thyroid problems than men and nearly half of all women and a quarter of all men in the US will die with evidence of an inflamed thyroid.
• The symptoms of this disease often vary from person to person and are non-specific, so a correct diagnosis can easily be missed or misdiagnosed for irrelevant issues.

Objective:

• Finding an accurate solution to this problem is a must
• Provide an efficient solution for healthcare practitioners via Support Vector Machines for diagnosing/classifying a particular thyroid disease that a person may have.
• This tool will cause an immense decrease in misdiagnoses as it is capable of distinguishing between problems of the thyroid gland and other illnesses in the body.
• As well as providing the ability to detect the disease before it forms into a more destructive anomaly.

Thyroid: What is it?

The thyroid is a gland that produces thyroid hormone. This hormone regulates vital body functions:

• Breathing
• Body weight
• Heart rate
• Muscle Strength

Types of Thyroid Diseases

• Euthyroid: A normal functioning thyroid gland

• Euthyroid Sick: The serum levels of thyroid hormones are low in clinically euthyroid patients with nonthyroidal systemic illness

• Hypothyroidism: The thyroid fails to make enough thyroid hormones and slows down many of your body's functions

• Hyperthyroidism: The thyroid makes more thyroid hormone than the body needs, which speeds up many of your body's functions.

Thyroid Data Preprocessing: Exploratory Analysis

• Goal: gain a solid understanding of the data and make corrections before training model
• Imputation of missing values
  • For Continuous values: substituted in with the mode of the column
    • Mode is the most frequently occurring number Robust to outliers
  • For Categorical values (Sex): wasn’t many missing, so deleted those rows
    • Why? Because the sex of the person matters in respect to their diagnosis
• Balanced data: leaned heavily towards the negative samples
• Once found, able to train model

Data can be found here: https://archive.ics.uci.edu/ml/datasets/thyroid+disease

Support Vector Machines

• A geometric approach to classification and regression tasks
• Main Objective: to find a hyperplane that distinctly classifies the data points.
• Hyperplanes are decision boundaries that help classify the data points.
• If a data point falls on either side of the hyperplane, it is attributed to different classes. 
• There exist many possible hyperplanes, but the one that has the maximum margin, or distance, between data points of the classes is ideal.
• This maximum marginal distance provides confidence that future data points are assigned to the correct class

One-vs-Rest Scheme

• SVMs are typically used for binary classification; classifying between two classes
• Strategy involves training a single classifier per class, with the samples of that class as positive samples and all other samples as negatives. 
• Essentially, it decomposes a multiclass classification problem into a multiple binary classification problems

Results:

• Obtained an accuracy of about 69%
• For further details on the entire process, visit the notebook here

Required Packages

Python Version: 3.6.5

Pandas: Powerful data structures for data analysis, time series, and statistics

pip install pandas

Numpy: a general-purpose array-processing package designed to efficiently manipulate large multi-dimensional arrays

pip install numpy

Scikit-Learn: Simple and efficient tools for data mining and data analysis

pip install scikit-learn

Seaborn: A library for making statistical graphics in Python

pip install seaborn

Matplotlib: Python plotting package

pip install matplotlib