Diabetes-Prediction-Using-KNN-Algorithm_From_Scratch

Supervised Problem:

• The problem in which we are given a dataset which has target/dependent/labelled variable.
• Here we know for what we are building the model

Classification Problem:

• The problem where dataset contains categorical target variable.

KNN

• K Nearest Neighbours
• Uses similarity measure ( how much two objects are alike)
• Stores available cases and for finding label for new case, check in it's v neighbor, and says that I am one of them.
• Applications

1. Recommended System (E Commerce Websites)
2. Concept Search (Internet generates plethora of documents each day, to segeregate them we may use this algorithm)

My Implementation

• Algorithm

1. Find the distance between new_data_instance and all existing instances
2. Find nearest K points.
3. Among them choose which target class occurred in majority.

Catch

1. Choose approaprite value for K

Steps

1. Define Distance Metric Function (Euclidian & Manhattan)
2. Define NearestNeigbours Function
3. Define Predict Function

• X_train --> training data with features and target

• X_test --> test data without target

• K: K neighbors

Summary

• Total Attributes - 9
• Number of instances - 768
• Score (Accuracy) - 74.4%
• View Notebook