The comparison of a Decision Tree Model and a K-Nearest Neighbors model on the determination of Diabetic Retinopathy

Anish VermaFollow

School Name

Spring Valley High School

Grade Level

11th Grade

Presentation Topic

Computer Science

Presentation Type

Non-Mentored

Abstract

Diabetic retinopathy is recognized as a leading cause of vision impairment, attributed to the distortion and scarring of retinal blood vessels (Bilal et al., 2021). The worldwide impact of diabetic retinopathy is on the rise, contributing to its status as a prominent cause of vision impairment globally (Arcadu et al., 2019). This research explores the comparative analysis of two machine learning models, the Decision Tree and K-Nearest Neighbors, to discover which model has a more accurate performance in classifying diabetic retinopathy severity levels. It was hypothesized that if the Decision Tree and K-Nearest Neighbors models were used to classify images of Diabetic Retinopathy, then Decision Tree would have a more accurate classification. Using a Kaggle dataset of retinal pathology images, the InceptionV3 convolutional neural network was employed for feature extraction. NumPy and pandas handle dataset manipulation, and the InceptionV3 model, excluding top layers, extracted features for hybrid integration with the Decision Tree and K-Nearest Neighbors models. The exclusion of top layers focuses on feature extraction rather than image classification, enhancing accuracy. The Decision Tree, a model renowned for its interpretability and simplicity, is pitted against the KNN model, known for its reliance on proximity-based classification. The results of this experiment indicated that there was a significant difference in the accuracy between the Decision Tree classifier and the K-Nearest Neighbors classifier. This was due to the unpaired t-test which obtained a two-tailed P value measuring less than 0.0001 signifying an extremely statistically significant difference.

Location

RITA 367

Start Date

3-23-2024 10:15 AM

Presentation Format

Oral and Written

Group Project

No

COinS
 
Mar 23rd, 10:15 AM

The comparison of a Decision Tree Model and a K-Nearest Neighbors model on the determination of Diabetic Retinopathy

RITA 367

Diabetic retinopathy is recognized as a leading cause of vision impairment, attributed to the distortion and scarring of retinal blood vessels (Bilal et al., 2021). The worldwide impact of diabetic retinopathy is on the rise, contributing to its status as a prominent cause of vision impairment globally (Arcadu et al., 2019). This research explores the comparative analysis of two machine learning models, the Decision Tree and K-Nearest Neighbors, to discover which model has a more accurate performance in classifying diabetic retinopathy severity levels. It was hypothesized that if the Decision Tree and K-Nearest Neighbors models were used to classify images of Diabetic Retinopathy, then Decision Tree would have a more accurate classification. Using a Kaggle dataset of retinal pathology images, the InceptionV3 convolutional neural network was employed for feature extraction. NumPy and pandas handle dataset manipulation, and the InceptionV3 model, excluding top layers, extracted features for hybrid integration with the Decision Tree and K-Nearest Neighbors models. The exclusion of top layers focuses on feature extraction rather than image classification, enhancing accuracy. The Decision Tree, a model renowned for its interpretability and simplicity, is pitted against the KNN model, known for its reliance on proximity-based classification. The results of this experiment indicated that there was a significant difference in the accuracy between the Decision Tree classifier and the K-Nearest Neighbors classifier. This was due to the unpaired t-test which obtained a two-tailed P value measuring less than 0.0001 signifying an extremely statistically significant difference.