A Study of Disease Diagnosis using

Machine Learning
Samin Poudel*
*ComputationalData Science and Engineering, North Carolina A&T State University,
Greensboro, NC 27409

1
Presentation follows as below:

➢ Introduction
➢ Data, Algorithms and Methods
➢ Result and Discussion
➢ Conclusion and Future Work

2
 Introduction

• Artificial Intelligence (AI)

and Machine Learning (ML)
is successfully applied to
practically in every domain
like robotics, education,
travel to health care
• Various applications of ML
in healthcare as shown in
the Figure 1
• In this past decade, the
investment in AI in Figure 1: Applications of Machine Learning in Healthcare*
healthcare applications has
increased significantly
3

*https://data-flair.training/blogs/machine-learning-in-healthcare/
 Introduction

• The analysis of the clinical data can lead to the timely diagnosis of the
disease which will help to start cure for the patient in time as well
• Traditional approach of diagnosing disease is generally costly and time
consuming
• ML techniques have not only been able to diagnose the common
diseases but are also equally capable of diagnosing the rare diseases
• In general, a dataset table used to build a ML model for diagnosing a
disease have columns for different attributes and a column variable for
the class variable

4
 Introduction
Problem Statement:
• Accuracy of the ML in diagnosing the diseases is still a concern
• Improvement in the performance of ML to diagnose disease is a hot topic in
healthcare domain
• Different ML approach perform differently for different healthcare dataset
• Need to find the way to apply many state of art algorithms to same dataset
in reasonable time with minimal lines of codes, so that the search of best
ML method can be pursued efficiently to diagnose a particular disease

Probable Solution:
• The use of libraries like AutoGluon can help to find the best performing ML
approach out of many ML approaches in diagnosing the disease for a given
dataset with optimal lines of codes.
5
 Data, Algorithms and Methods
Data:
• Dataset Used: Pima Indian Diabetes
• This data set has 8 attributes and one class variable named Outcome.
• Outcome variable has value of 0 or 1, 1 means tested positive for diabetes
• The dataset has 768 instances, 268 instances are tested positive for diabetes
Table 1. Statistical description of Data based on Attributes
Diabetes
Pregnancies Glucose Blood Pressure Skin Thickness Insulin BMI Pedigree Age
Function

Count 768 768 768 768 768 768 768 768

Mean 3.85 120.89 69.10 20.57 79.79 31.99 0.47 33.24

std 3.37 31.97 19.35 15.95 115.244 7.88 0.33 11.76

min 0 0 0 0 0 0 0.078 21

25% (Q1) 1 99 62 0 0 27.3 0.24 24

50% (Q2) 3 117 72 23 30.5 32 0.37 29

75% (Q3) 6 140.25 80 32 127.25 36.6 0.63 41 6

max 17 199 122 99 846 67.1 2.42 81.0

 Data, Algorithms and Methods
Data:
• Data Exploratory
Visualization
showed that ML
models can be built
without
preprocessing of
the data
• Every attribute may
be important for the
disease diagnosis
with Machine
Learning Figure 2: Histogram of Attributes
7
 Data, Algorithms and Methods
Machine Learning Algorithms Used:
• 20 Machine Learning Algorithms are used by importing from scikit-learn and
AutoGluon Libraries in AWS SageMaker
Table 2. List of ML Algorithms Used

Library Number of ML
ML Algorithm
approaches
Scikit-Learn Random Forest Classifier, Decision Tree Classifier, Naïve Bayes Classifier, Perceptron, 6
Multilayer Perceptron, Voting Classifier

WeightedEnsemble_L2, LightGBM_BAG_L1, LightGBM_LARGE_BAG_L1,

NeuralNetFastAI_BAG_L1, CATBoost_BAG_L1, ExtraTreesGini_BAG_L1,
AutoGluon LightGBMXT_BAG_L1, XGBoost_BAG_L1, RandomForestEntr_BAG_L1, 14
RandomForestGini_BAG_L1, ExtraTreesEntr_BAG_L1, NeuralNetMXNet_BAG_L1,
KNeighborsUnif_BAG_L1, KNeighborsDist_BAG_L1

8
 Data, Algorithms and Methods

• Overview of Methodology:
• Data Loaded to Amazon SageMaker’s Jupyter Instance
• Data Spitted to Training and Test set
• Machine Learning Algorithms trained and tested using scikit-learn and
AutoGluon Library
• Training and Test set for each of the ML algorithm should be same for
reasonable comparable among them. It was achieved by defining
random seed while splitting data into training and test sets
• Evaluation of ML algorithms to diagnose diabetes are performed using
classification metrics Accuracy, Precision, Recall and F1-score
• Detailed Implementation of the ML algorithms is in authors’ GitHub page
9
 Result and Discussion
Evaluation of ML Algorithms:
• Although being a classical ML algorithm, Naive Bayes performed better among
the ML algorithms, based on combined analysis of all the evaluation metrics

Table 3. Evaluation of ML Algorithms

S. N ML Algorithm Accuracy F1-score Precision Recall

1 Random Forest Classifier (Scikit-learn) 0.74 0.81 0.78 0.84

2 Decision Tree Classifier (Scikit-learn) 0.65 0.73 0.73 0.73
3 Naïve Bayes Classifier (Scikit-learn) 0.77 0.83 0.80 0.86
4 Perceptron (Scikit-learn) 0.49 0.47 0.71 0.35
5 Multilayer Perceptron (Scikit-learn) 0.68 0.76 0.75 0.77
6 Voting Classifier (Scikit-learn) 0.72 0.78 0.79 0.77
7 AutoGluon Best Performer 0.74 0.82 0.76 0.88
10
 Result and Discussion
• Accuracy performance of different AutoGluon ML algorithms when trained with accuracy as validation
metric is in Figure 3a. Similarly, performance in terms of F1-scores is shown when trained with F1-scores
as validation metric in Figure 3b
• Weighted Ensemble_L2 ML technique performs better for both the cases and KNN based ML has the least
performance for both the cases

Figure 3. (a) Evaluation of AutoGluon ML algorithms when trained with accuracy as validation metric 11

(b) Evaluation of AutoGluon ML algorithms when trained with F1-score as validation metric
 Conclusion and Future Work

Conclusion:
• Libraries like AutoGluon help comparing performances of many ML approaches in diagnosing a
disease for a given dataset with optimal lines of code.
• This helps in finding the best performing ML algorithm for a particular dataset or a particular
type of disease as well. And it decreases the probability of inaccurate diagnosis, which is a
significantly important consideration while dealing with the health of the people.
• Performance of 20 ML approaches in diagnosing diabetes based on the Pima Indian Diabetes
Dataset tested
• For the data set considered, Naïve Bayes algorithm performed better among the other
algorithms. This shows that using the complex and computationally costly algorithms not
necessarily improve the accuracy of diagnosing a disease.

Future Work:
• The possibility of the improvement in the performance of ML models in future can be started by
finding the correlation among each attribute and dropping the highly correlated attributes.
Because the highly correlated attributes can confuse a model in the learning phase. 12
 Thank you

THANK YOU

13