search

sharinkan / HeartDiseasePrediction

0 stars 0 forks source link

readme

Deep Learning: Heart Disease Prediction

TeamName: CardioWave Forecasters

Instructors: Milad Toutounchian, Mahmoud Zeydabadinezhad

Group members: Ziao You, Weijie Chen, Xiao Fang, Yixuan Song, Kan Kim

Email Address: zy364@drexel.edu

#

Introduction

In the field of healthcare, the stethoscope has been an indispensable tool for diagnosing cardiovascular conditions by listening to the heart's sounds. These auditory cues provide valuable information to skilled medical professionals and we are attempting to use machine learning to give an early indication of heart conditions

Methods

Machine learning models

Deep learning models

Results (DL) Updated by 06/05/2024

Model Features Accuracy
Single MLP Chromagram + 1D Mel-Spec + MFCC + Opensmile + CSV + Time Series 66.46
Multiscale 2D CNN 2D Mel-Spec 58.23
1D CNN Bandpower Struct + CSV 68.95
Concat-CNN Chromagram + 1D Mel-Spec + MFCC + Bandpower Struct + CSV 64.22
Conv-RNN 1D Mel-Spec + MFCC + Opensmile + CSV 64.38
ResNet Chromagram + 1D Mel-Spec + MFCC + Time Series 55.22
Concat-MLP 2D Mel-Spec + Bandpower Struct + Opensmile + CSV 59.11
LSTM Chromagram + 2D Mel-Spec + CSV + Time Series 53.85
RNN Chromagram + Bandpower Struct + CSV + Time Series 50.00

Results (ML)

Different Input Features + Same ML Models + Voting

['mfcc'] + hard voting
                   Model  Accuracy   AUC
0      LogisticRegression  0.633987  0.64
1                     SVC  0.627451  0.63
2    KNeighborsClassifier  0.568627  0.57
3  DecisionTreeClassifier  0.526144  0.52
4  RandomForestClassifier  0.584967  0.59
5              GaussianNB  0.594771  0.60
ensemble_methods : VOTING='hard' 0.630718954248366

['mfcc'] + soft voting
                   Model  Accuracy   AUC
0      LogisticRegression  0.545752  0.55
1                     SVC  0.581699  0.59
2    KNeighborsClassifier  0.526144  0.53
3  DecisionTreeClassifier  0.555556  0.56
4  RandomForestClassifier  0.558824  0.57
5              GaussianNB  0.542484  0.54
ensemble_methods : VOTING='soft' 0.545751633986928

['feature_chromagram'] + hard voting
                   Model  Accuracy   AUC
0      LogisticRegression  0.578431  0.58
1                     SVC  0.555556  0.56
2    KNeighborsClassifier  0.549020  0.55
3  DecisionTreeClassifier  0.496732  0.50
4  RandomForestClassifier  0.535948  0.54
5              GaussianNB  0.529412  0.52
ensemble_methods : VOTING='hard' 0.5686274509803921

['feature_chromagram'] + soft voting

                   Model  Accuracy   AUC
0      LogisticRegression  0.532680  0.53
1                     SVC  0.526144  0.53
2    KNeighborsClassifier  0.509804  0.51
3  DecisionTreeClassifier  0.558824  0.56
4  RandomForestClassifier  0.513072  0.51
5              GaussianNB  0.500000  0.50
ensemble_methods : VOTING='soft' 0.5294117647058824

['feature_melspectrogram'] + hard voting
                   Model  Accuracy   AUC
0      LogisticRegression  0.575163  0.56
1                     SVC  0.555556  0.54
2    KNeighborsClassifier  0.575163  0.57
3  DecisionTreeClassifier  0.522876  0.52
4  RandomForestClassifier  0.571895  0.57
5              GaussianNB  0.539216  0.52
ensemble_methods : VOTING='hard' 0.565359477124183

['feature_melspectrogram'] + soft voting

                   Model  Accuracy   AUC
0      LogisticRegression  0.552288  0.52
1                     SVC  0.549020  0.50
2    KNeighborsClassifier  0.526144  0.53
3  DecisionTreeClassifier  0.509804  0.50
4  RandomForestClassifier  0.584967  0.59
5              GaussianNB  0.568627  0.51
ensemble_methods : VOTING='soft' 0.5620915032679739

[feature_bandpower]
                   Model  Accuracy   AUC
0      LogisticRegression  0.496732  0.51
1                     SVC  0.516340  0.53
2    KNeighborsClassifier  0.532680  0.53
3  DecisionTreeClassifier  0.519608  0.52
4  RandomForestClassifier  0.539216  0.54
5              GaussianNB  0.486928  0.52
ensemble_methods : VOTING='hard' 0.5065359477124183

[NMF]
                   Model  Accuracy   AUC
0      LogisticRegression  0.493464  0.49
1                     SVC  0.562092  0.56
2    KNeighborsClassifier  0.500000  0.50
3  DecisionTreeClassifier  0.480392  0.48
4  RandomForestClassifier  0.519608  0.52
5              GaussianNB  0.535948  0.51
ensemble_methods : VOTING='hard' 0.5392156862745098

['feature_mfcc', 'feature_chromagram']

                   Model  Accuracy   AUC
0      LogisticRegression  0.594771  0.60
1                     SVC  0.584967  0.59
2    KNeighborsClassifier  0.496732  0.50
3  DecisionTreeClassifier  0.539216  0.54
4  RandomForestClassifier  0.535948  0.54
5              GaussianNB  0.562092  0.56
ensemble_methods : VOTING='hard' 0.5686274509803921

[feature_mfcc, feature_bandpower]

                   Model  Accuracy   AUC
0      LogisticRegression  0.565359  0.57
1                     SVC  0.601307  0.60
2    KNeighborsClassifier  0.496732  0.49
3  DecisionTreeClassifier  0.555556  0.56
4  RandomForestClassifier  0.549020  0.55
5              GaussianNB  0.503268  0.51
ensemble_methods : VOTING='hard' 0.5751633986928104

Different Input Features + Different ML Models

Voting (hard)
                        Model  Accuracy   AUC
0                 LogisticRegression()_f1  0.594771  0.59
1                                SVC()_f1  0.591503  0.59
2                         GaussianNB()_f1  0.581699  0.58
3             RandomForestClassifier()_f1  0.539216  0.54
4                                SVC()_f2  0.486928  0.49
5  KNeighborsClassifier(n_neighbors=7)_f3  0.539216  0.54
6                 LogisticRegression()_f3  0.549020  0.55
ensemble_methods : VOTING='hard' 0.6111111111111112

Voting (soft)
                        Model  Accuracy   AUC
0      LogisticRegression()_f1  0.614379  0.61
1                     SVC()_f1  0.617647  0.62
2              GaussianNB()_f1  0.565359  0.57
3  RandomForestClassifier()_f1  0.604575  0.60
4      LogisticRegression()_f2  0.490196  0.49
5      LogisticRegression()_f3  0.549020  0.55
ensemble_methods : VOTING='soft' 0.5490196078431373

# Google Colab Folder: https://drive.google.com/drive/folders/1R7UYthWO2DxDS4M7F3t_E6CDGoRBOWn4?usp=share_link

# Dataset link: https://physionet.org/content/circor-heart-sound/1.0.3/

Setup script instruction

Python version : 3.9+

#

Install required packages

pip install -r requirements.txt #

Install dataset

python install.py

Training

python run.py