Diagnosing Diseases using kNN

Introduction

• The k-Nearest-Neighbors (kNN) is an algorithm that can be used in a variety of fields to classify or predict data (Ali et al. 2020)

• It’s a simple algorithm that classifies data based on how similar a datapoint is to a class of datapoints. (Zhang 2016)

• One of the benefits of using this algorithmic model is how simple it is to use and the fact it’s non-parametric which means it fits a wide variety of datasets.

• One drawback from using this model is that it does have a higher computational cost than other models which means that it doesn’t perform as well or fast on big data (Deng et al. 2016)

• In this project we focused on the methodology and application of classification kNN models in the field of healthcare or public health to predict or screen for diabetes.

Methodology Overview

• The kNN algorithm is a nonparametric supervised learning algorithm that can be used for classification or regression problems. (Syriopoulos et al. 2023)

The classification process has three distinct steps:

  K=5  vs K=15

  Majority wins 

Methodology Visualization

• Figure 1 illustrates this methodology with two distinct classes of hearts and circles.

Assumptions

• The kNN algorithm assumes similar datapoints will be in close proximity to each other and be neighbors (Zhang 2016).

• It also assumes that data points with similar features belong to the same class. (Boateng, Otoo, and Abaye 2020)

Pre-processing Data

• Handle missing values: We must remove the missing values by either inputting them or dropping them to prevent them from skewing the results.
• Make all values numeric: All categorical values must be encoded using either one-hot encoding or label encoding.
• Normalize or Standardize the features: We must use min-max scaler or the standard scaler to make sure we reduce bias.
• Reduce dimensionality: We can use Principal Component Analysis to reduce the number of features but keep the variance.
• Remove correlated features: The kNN works best when there aren’t too many features, so we can use a correlation matrix to see which features we can drop.
• Fix class imbalance: Synthetic Minority Over-sampling Technique(SMOTE) can be used to handle class imbalances that can cause biases.

Hyperparameter Tuning

In order to increase the accuracy of the model there are a few parameters that we can adjust.

1. Find the optimal k parameter: We can use gridsearch to find the best parameter k.

2. Change the distance metric: The kNN uses the euclidean distance by default but we can use the Manhattan distance, the Minkowski distance or another distance.

3. Weights: The kNN defaults to a “uniform” weight where it gives the same weight to all the distances but it can be adjusted to “distance” so that the closest neighbors have more weight.

Dataset Overview

• We explored the CDC Diabetes Health Indicators dataset, sourced from the UC Irvine Machine Learning Repository.

• The dataset consists of 253,680 survey responses and contains 21 feature variables and 1 binary target variable named Diabetes_binary

• Diabetes_binary: 0= No Diabetes, 1= Diabetes

• Binary Variables: HighBP, HighChol, CholCheck, Smoker, Stroke, HeartDiseaseorAttack, PhysActivity, Fruits, Veggies, HvyAlcoholConsump, AnyHealthcare, NoDocbcCost, DiffWalk, Sex.

• Ordinal Variables: GenHlth, MentHlth, PhysHlth, Age, Education, Income

• Continuous Variables: BMI

Data Exploration and Visualization - Outliers

Figure 5 shows us outliers in the data that can skew our results

Data Exploration and Visualization - Class imbalance

Figure 6 shows the class imbalance present in the data

Data Exploration and Visualization - Key Findings

• There are no missing values, meaning no imputation is needed.

• We have some duplicate values that need to be removed.

• There is a class imbalance with the majority of cases not having diabetes.

Building the Models

• There was no missing data so we didn’t have to remove or impute any values but we did clean the data by dropping duplicate values.
• We kept the ordinal variables the same as they have meaningful natural order that will provide the kNN with meaningful distances.
• The data was divided into testing and training data. We used a test_size=0.2 to use 80% of the data for training the kNN and 20% of the data for testing.
• The features were standardized so that BMI and age could be on the same scale as the other features.
• Gridsearch was used to find the optimal parameters for hyperparameter tuning
• We experimented with different decision thresholds due to the imbalanced dataset.

Modeling and Results - Model Creation

We chose to create four classification kNN models to illustrate the methodology.

Modeling and Results- Evaluating the models

The table below shows the summary of the four models.

Results

• Model 2 has the highest accuracy at 84.56% but this accuracy score is high because it is good at detecting the non-diabetic cases which are the majority of cases.

• Model 2 also has the highest ROC AUC score of 0.77 which means it’s the best model at separating different classes; however, the recall is 14.56% so it’s only classifying 14.56% of the actual positive diabetic cases.

• Model 3 has an accuracy of 69.77% and a much higher recall of 69.58%. Model 3 is able to correctly classify about 70% of the positive diabetic cases.

• Model 4 has the best recall with the ability to classify 91.62% of the positive diabetic cases but the overall accuracy is at 47.52% so it is classifying non-diabetic cases as diabetic.

Conclusion

• kNN is a promising algorithmic model that can be further improved to classify or screen for diabetes.

• Model 3 showed potential with classifying diabetic cases but would need to be further improved by being trained with data that shows more diabetic cases if it’s going to be used in a healthcare setting.

• Model 4 showed high potential for screening for diabetes and would be useful in a public health setting where people are classified as high risk for diabetes based on their self reported helath indicators and referred for lifestyle education and biochemical testing.