Assignment 2. Supervised Learning. AI.

Final Delivery

• David Burchakov ([email protected])

To run the program, open Project_2_Supervised_Learning.ipynb file with Jupyter Notebook.

Description

To see a more detailed description and the result, open the .ipynb file

Application of different Machine Learning algorithms, including Nearest Neighbor, SVM, Decision Trees and Neural Networks and a comparison of theese algorithms. The purpose of the project is to find a binary solution for real-life data - Regensburg Pediatric Appendicitis dataset. Machine Learning algorithms are taught to identify, whether an entry will have apendicitis or not based on his characteristics.

Steps

I. Correct the data

Firstly, the data should be corrected from erroneous data, empty entries or outliners.

Remove outliers

*outlier

Remove erroneous data

# - remove anomally tall babies
rows_to_drop = data[(data['Age'] <= 3.0) & (data['Height'] > 100)].index
dataNew = data.drop(rows_to_drop)

# - remove anomally cold patients
rows_to_drop = dataNew[(dataNew['Body_Temperature'] <= 30)].index
dataNew = dataNew.drop(rows_to_drop)

# - remove anomally big RDW
rows_to_drop = dataNew[(dataNew['RDW'] > 30)].index
dataNew = dataNew.drop(rows_to_drop)

# Remove anomaly heavy children (aged less than 3 years)
rows_to_drop = dataNew[(dataNew['Age'] < 2) & (dataNew['Weight'] > 14.8)].index
dataNew = dataNew.drop(rows_to_drop)

Additionally, we remove columns with incremental unique unformation that does not affect the final binary solution

US_number is an incremental unique number of the patient that does not influence having appendicitis or not.

II. Analyse the data

In order to anylise the data, we can make a Heatmap graph that indicates the correlation between each characteristic (column) and desired result (having appendicitis: yes/no) and drop those columns that do not affect the result.

III. Apply Machine Learning Algorithms

Algorithms applied:

a) Nearest Neighbor

b) SVMs

c) Decision Trees

d) Neural Network