Прогнозирование сердечных заболеваний UCI с использованием NN

Контекст

Эта база данных содержит 76 атрибутов, но все опубликованные эксперименты относятся к подмножеству из 14 из них. В частности, база данных Кливленда - единственная, которая
использовалась исследователями машинного обучения к этой дате. Поле «цель» относится к наличию у пациента сердечного заболевания. Это целочисленное значение от 0 (отсутствие присутствия) до 4.

Содержание

Информация об атрибутах:

возраст
секс
тип боли в груди (4 значения)
артериальное давление в покое
холестерин сыворотки в мг / дл
уровень сахара в крови натощак ›120 мг / дл
результаты электрокардиографии в покое (значения 0,1,2)
достигнута максимальная частота сердечных сокращений
стенокардия, вызванная физической нагрузкой
старый пик = депрессия ST, вызванная упражнениями по сравнению с отдыхом
наклон сегмента ST при пиковой нагрузке
количество крупных сосудов (0–3), окрашенных при рентгеноскопии
thal: 3 = нормальный; 6 = исправленный дефект; 7 = обратимый дефект

Импорт библиотек

Прочитать данные

загрузка данных в блокнот

Анализ данных

Визуализация данных с помощью графиков и тепловых карт

Предварительная обработка данных

Создание фиктивных столбцов

Index(['age', 'sex', 'trestbps', 'chol', 'fbs', 'restecg', 'thalach', 'exang',
       'oldpeak', 'ca', 'target', 'cp_0', 'cp_1', 'cp_2', 'cp_3', 'thal_0',
       'thal_1', 'thal_2', 'thal_3', 'slope_0', 'slope_1', 'slope_2'],
      dtype='object')

age           int64
sex           int64
trestbps      int64
chol          int64
fbs           int64
restecg       int64
thalach       int64
exang         int64
oldpeak     float64
ca            int64
target        int64
cp_0          uint8
cp_1          uint8
cp_2          uint8
cp_3          uint8
thal_0        uint8
thal_1        uint8
thal_2        uint8
thal_3        uint8
slope_0       uint8
slope_1       uint8
slope_2       uint8
dtype: object

Колонна сброса отходов

Поезд-тестовый сплит

Разделение набора данных на 80–20 частей

Модель Архитектура

Архитектура с использованием нейронных сетей

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 12)                264       
_________________________________________________________________
dropout (Dropout)            (None, 12)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 2)                 26        
=================================================================
Total params: 290
Trainable params: 290
Non-trainable params: 0
_________________________________________________________________

Обучение модели

Обученная модель с использованием Adam Optimiser

Epoch 1/50
22/22 [==============================] - 0s 7ms/step - loss: 0.6330 - accuracy: 0.7311 - val_loss: 0.6195 - val_accuracy: 0.7143
Epoch 2/50
22/22 [==============================] - 0s 2ms/step - loss: 0.5644 - accuracy: 0.8349 - val_loss: 0.5751 - val_accuracy: 0.7253
Epoch 3/50
22/22 [==============================] - 0s 2ms/step - loss: 0.5021 - accuracy: 0.8679 - val_loss: 0.5326 - val_accuracy: 0.7363
Epoch 4/50
22/22 [==============================] - 0s 2ms/step - loss: 0.4519 - accuracy: 0.8774 - val_loss: 0.5008 - val_accuracy: 0.7802
Epoch 5/50
22/22 [==============================] - 0s 2ms/step - loss: 0.4095 - accuracy: 0.8679 - val_loss: 0.4799 - val_accuracy: 0.7692
Epoch 6/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3760 - accuracy: 0.8632 - val_loss: 0.4683 - val_accuracy: 0.7692
Epoch 7/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3642 - accuracy: 0.8726 - val_loss: 0.4642 - val_accuracy: 0.7692
Epoch 8/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3463 - accuracy: 0.8585 - val_loss: 0.4645 - val_accuracy: 0.7582
Epoch 9/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3448 - accuracy: 0.8821 - val_loss: 0.4648 - val_accuracy: 0.7692
Epoch 10/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3219 - accuracy: 0.8821 - val_loss: 0.4651 - val_accuracy: 0.7802
Epoch 11/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3218 - accuracy: 0.9009 - val_loss: 0.4669 - val_accuracy: 0.7802
Epoch 12/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3298 - accuracy: 0.8868 - val_loss: 0.4714 - val_accuracy: 0.7692
Epoch 13/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3189 - accuracy: 0.8868 - val_loss: 0.4727 - val_accuracy: 0.7802
Epoch 14/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3097 - accuracy: 0.8821 - val_loss: 0.4759 - val_accuracy: 0.7582
Epoch 15/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2999 - accuracy: 0.8962 - val_loss: 0.4785 - val_accuracy: 0.7802
Epoch 16/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3017 - accuracy: 0.8962 - val_loss: 0.4840 - val_accuracy: 0.7692
Epoch 17/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2890 - accuracy: 0.8915 - val_loss: 0.4835 - val_accuracy: 0.7802
Epoch 18/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2858 - accuracy: 0.8962 - val_loss: 0.4903 - val_accuracy: 0.7582
Epoch 19/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2978 - accuracy: 0.8962 - val_loss: 0.4915 - val_accuracy: 0.7692
Epoch 20/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3000 - accuracy: 0.8868 - val_loss: 0.4947 - val_accuracy: 0.7802
Epoch 21/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2750 - accuracy: 0.8962 - val_loss: 0.4994 - val_accuracy: 0.7692
Epoch 22/50
22/22 [==============================] - 0s 2ms/step - loss: 0.3057 - accuracy: 0.8962 - val_loss: 0.5014 - val_accuracy: 0.7802
Epoch 23/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2841 - accuracy: 0.8962 - val_loss: 0.5023 - val_accuracy: 0.7692
Epoch 24/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2750 - accuracy: 0.8962 - val_loss: 0.5018 - val_accuracy: 0.7692
Epoch 25/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2843 - accuracy: 0.9151 - val_loss: 0.5044 - val_accuracy: 0.7692
Epoch 26/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2800 - accuracy: 0.8868 - val_loss: 0.5091 - val_accuracy: 0.7692
Epoch 27/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2852 - accuracy: 0.9009 - val_loss: 0.5023 - val_accuracy: 0.7802
Epoch 28/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2617 - accuracy: 0.9009 - val_loss: 0.5051 - val_accuracy: 0.7692
Epoch 29/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2692 - accuracy: 0.8962 - val_loss: 0.5066 - val_accuracy: 0.7802
Epoch 30/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2681 - accuracy: 0.9009 - val_loss: 0.5142 - val_accuracy: 0.7692
Epoch 31/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2709 - accuracy: 0.9009 - val_loss: 0.5110 - val_accuracy: 0.7802
Epoch 32/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2801 - accuracy: 0.8868 - val_loss: 0.5114 - val_accuracy: 0.7912
Epoch 33/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2667 - accuracy: 0.8868 - val_loss: 0.5131 - val_accuracy: 0.7912
Epoch 34/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2741 - accuracy: 0.8868 - val_loss: 0.5107 - val_accuracy: 0.7802
Epoch 35/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2660 - accuracy: 0.9151 - val_loss: 0.5130 - val_accuracy: 0.7802
Epoch 36/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2781 - accuracy: 0.9009 - val_loss: 0.5132 - val_accuracy: 0.7802
Epoch 37/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2711 - accuracy: 0.9151 - val_loss: 0.5154 - val_accuracy: 0.7692
Epoch 38/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2584 - accuracy: 0.9104 - val_loss: 0.5148 - val_accuracy: 0.7802
Epoch 39/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2726 - accuracy: 0.8962 - val_loss: 0.5170 - val_accuracy: 0.7802
Epoch 40/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2684 - accuracy: 0.8962 - val_loss: 0.5225 - val_accuracy: 0.7802
Epoch 41/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2590 - accuracy: 0.9057 - val_loss: 0.5195 - val_accuracy: 0.7802
Epoch 42/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2608 - accuracy: 0.8962 - val_loss: 0.5241 - val_accuracy: 0.7802
Epoch 43/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2537 - accuracy: 0.9151 - val_loss: 0.5263 - val_accuracy: 0.7802
Epoch 44/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2485 - accuracy: 0.8962 - val_loss: 0.5251 - val_accuracy: 0.7802
Epoch 45/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2368 - accuracy: 0.9151 - val_loss: 0.5270 - val_accuracy: 0.7802
Epoch 46/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2573 - accuracy: 0.8962 - val_loss: 0.5306 - val_accuracy: 0.7802
Epoch 47/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2395 - accuracy: 0.9057 - val_loss: 0.5328 - val_accuracy: 0.7802
Epoch 48/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2634 - accuracy: 0.9104 - val_loss: 0.5322 - val_accuracy: 0.7802
Epoch 49/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2447 - accuracy: 0.9151 - val_loss: 0.5346 - val_accuracy: 0.7802
Epoch 50/50
22/22 [==============================] - 0s 2ms/step - loss: 0.2518 - accuracy: 0.9057 - val_loss: 0.5403 - val_accuracy: 0.7802

Модели точности и потерь

Результат и заключение

precision    recall  f1-score   support
           0       0.78      0.71      0.74        41
           1       0.78      0.84      0.81        50
   micro avg       0.78      0.78      0.78        91
   macro avg       0.78      0.77      0.78        91
weighted avg       0.78      0.78      0.78        91
 samples avg       0.78      0.78      0.78        91

Сохранение модели

array([1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 0, 1, 1, 0, 1, 0, 0, 0, 1, 0, 1, 1,
       1, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 0, 1, 1, 1, 1, 0, 1,
       1, 0, 0, 0, 0, 1, 1, 0, 1, 1, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 1, 0,
       1, 0, 0, 1, 1, 1, 1, 1, 1, 0, 1, 1, 0, 0, 1, 0, 1, 1, 1, 0, 1, 1,
       1, 0, 0])

DeepCC

[INFO]
Reading [keras model] 'heart_disease_ucl.h5'
[SUCCESS]
Saved 'heart_disease_ucl_deepC/heart_disease_ucl.onnx'
[INFO]
Reading [onnx model] 'heart_disease_ucl_deepC/heart_disease_ucl.onnx'
[INFO]
Model info:
  ir_vesion : 4
  doc       : 
[WARNING]
[ONNX]: terminal (input/output) dense_input's shape is less than 1. Changing it to 1.
[WARNING]
[ONNX]: terminal (input/output) dense_1's shape is less than 1. Changing it to 1.
WARN (GRAPH): found operator node with the same name (dense_1) as io node.
[INFO]
Running DNNC graph sanity check ...
[SUCCESS]
Passed sanity check.
[INFO]
Writing C++ file 'heart_disease_ucl_deepC/heart_disease_ucl.cpp'
[INFO]
deepSea model files are ready in 'heart_disease_ucl_deepC/' 
[RUNNING COMMAND]
g++ -std=c++11 -O3 -fno-rtti -fno-exceptions -I. -I/opt/tljh/user/lib/python3.7/site-packages/deepC-0.13-py3.7-linux-x86_64.egg/deepC/include -isystem /opt/tljh/user/lib/python3.7/site-packages/deepC-0.13-py3.7-linux-x86_64.egg/deepC/packages/eigen-eigen-323c052e1731 "heart_disease_ucl_deepC/heart_disease_ucl.cpp" -D_AITS_MAIN -o "heart_disease_ucl_deepC/heart_disease_ucl.exe"
[RUNNING COMMAND]
size "heart_disease_ucl_deepC/heart_disease_ucl.exe"
   text	   data	    bss	    dec	    hex	filename
 122165	   2984	    760	 125909	  1ebd5	heart_disease_ucl_deepC/heart_disease_ucl.exe
[SUCCESS]
Saved model as executable "heart_disease_ucl_deepC/heart_disease_ucl.exe"

Вот как мы создаем модель для прогнозирования набора данных о сердечных заболеваниях с помощью нейронных сетей. Существует миллион способов создать модель благодаря недавним и предстоящим разработкам в области глубокого обучения.

Ссылка на блокнот: Здесь

Кредиты: Сиддхарт Ганджу