基于机器学习的临床决策支持-ANN

声明：本文示例来自于GitHub用户vkasojhaa的项目，一切权利归其所有，此处仅是自己学习分享。

实现了基于机器学习的乳腺癌的恶性和良性预测，比较了不同机器学习算法之间的性能。主要目的是评估在每种算法的准确性和效率方面对数据进行分类的正确性。

loss

# 损失值：预估值与实际值之间的均方差

optimizer

# 优化器

trainer = optimizer.minimize(loss)

# 训练：最小化损失函数

基于机器学习(ANN)的乳腺癌预测

代码示例

       #导入依赖库
       #!/usr/bin/python3
       from __future__ import print_function
       import keras
       from keras.datasets import mnist
       from keras.models import Sequential
       from keras.layers import Dense, Dropout, Flatten
       from keras.layers import Conv2D, MaxPooling2D
       from keras import backend as K
       import numpy as np
       import pandas as pd
       from matplotlib import pyplot as plt
       from IPython.display import clear_output
       from keras.utils import plot_model
   
  

       #载入数据并进行数据预处理
       data = pd.read_csv("data.csv")
       data.head()
       data.drop('id',axis=1,inplace=True)
       data.drop('Unnamed: 32',axis=1,inplace=True)
       data['diagnosis'] = data['diagnosis'].map({'M':1,'B':0})
       data.head()
   
  

print("Row, Col", data.shape)# (row,col)

  

(data['diagnosis'][:398]==1).sum(),(data['diagnosis'][:398]==0).sum()

  

(data['diagnosis'][398:]==1).sum(),(data['diagnosis'][398:]==0).sum()

  

       mat=data.as_matrix()
       mat.shape
       mat
   
  

模型训练：

       #Using 2 Sigmoid Layers and RMSprop optimizer
       model2 = Sequential()
       model2.add(Dense(500, activation='sigmoid', use_bias=True, input_shape=(30,)))
       model2.add(Dense(1, activation='sigmoid'))
       keras.optimizers.RMSprop(lr=0.01, rho=0.9, epsilon=None, decay=0.0)
       model2.compile(optimizer='rmsprop',loss='binary_crossentropy', metrics=['accuracy'])
       history01 = model2.fit(mat[:,1:],mat[:,0], validation_split=0.3,shuffle=False,epochs=3000, batch_size=128,verbose=0)
   
  

       score = model2.evaluate(mat[:398,1:],mat[:398,0], verbose=0, batch_size=128)
       print('Train loss:', score[0])
       print('Train accuracy:', score[1])
   
  

       Train loss: 0.00821199454791287
       Train accuracy: 1.0
   
  

       score = model2.evaluate(mat[398:,1:],mat[398:,0], verbose=0, batch_size=128)
       print('Validation loss:', score[0])
       print('Validation accuracy:', score[1])
   
  

       Validation loss: 0.1262894694568121
       Validation accuracy: 0.9473684154755888
   
  

       plt.plot(history01.history['acc'], label='acc')
       plt.plot(history01.history['val_acc'], label='val_acc')
       plt.legend()
       plt.show()
       plt.plot(history01.history['loss'], label='loss')
       plt.plot(history01.history['val_loss'], label='val_loss')
       plt.legend()
       plt.show()
       count=0
       for i in history01.history['acc']:
           if i>0.99:
               count+=1
       print(count)
   
  

       #Using 3 Sigmoid Layers and RMSprop optimizer
       model3 = Sequential()
       model3.add(Dense(500, activation='sigmoid', use_bias=True, input_shape=(30,)))
       model3.add(Dense(500, activation='sigmoid', use_bias=True))
       model3.add(Dense(1, activation='sigmoid'))
       keras.optimizers.RMSprop(lr=0.01, rho=0.9, epsilon=None, decay=0.0)
       model3.compile(optimizer='rmsprop',
                     loss='binary_crossentropy',
                     metrics=['accuracy'])
       history0 = model3.fit(mat[:,1:],mat[:,0], validation_split=0.3,shuffle=False,epochs=3000, batch_size=128, verbose=0)
   
  

       score2 = model3.evaluate(mat[:398,1:],mat[:398,0], verbose=0, batch_size=128)
       print('Train loss:', score2[0])
       print('Train accuracy:', score2[1])
   
  

       Train loss: 0.011901359889894986
       Train accuracy: 0.992462311557789
   
  

       score2 = model3.evaluate(mat[398:,1:],mat[398:,0], verbose=0, batch_size=128)
       print('Validation loss:', score2[0])
       print('Validation accuracy:', score2[1])
   
  

       Validation loss: 0.0804629116945448
       Validation accuracy: 0.9707602311296073
   
  

       plt.plot(history0.history['acc'], label='acc')
       plt.plot(history0.history['val_acc'], label='val_acc')
       plt.legend()
       plt.show()
       plt.plot(history0.history['loss'], label='loss')
       plt.plot(history0.history['val_loss'], label='val_loss')
       plt.legend()
       plt.show()
       count=0
       for i in history0.history['acc']:
           if i>0.99:
               count+=1
       print(count)
   
  

       #Using 3 Sigmoid Layers and RMSprop optimizer
       model3 = Sequential()
       model3.add(Dense(500, activation='sigmoid', use_bias=True, input_shape=(30,)))
       model3.add(Dense(500, activation='sigmoid', use_bias=True))
       model3.add(Dense(1, activation='sigmoid'))
       keras.optimizers.RMSprop(lr=0.01, rho=0.9, epsilon=None, decay=0.0)
       model3.compile(optimizer='rmsprop',
                     loss='binary_crossentropy',
                     metrics=['accuracy'])
       history0 = model3.fit(mat[:,1:],mat[:,0], validation_split=0.3,shuffle=False,epochs=3000, batch_size=128, verbose=0)
       score2 = model3.evaluate(mat[:398,1:],mat[:398,0], verbose=0, batch_size=128)
       print('Train loss:', score2[0])
       print('Train accuracy:', score2[1])
   
  

       Validation loss: 0.0804629116945448
       Validation accuracy: 0.9707602311296073
   
  

       plt.plot(history0.history['acc'], label='acc')
       plt.plot(history0.history['val_acc'], label='val_acc')
       plt.legend()
       plt.show()
       plt.plot(history0.history['loss'], label='loss')
       plt.plot(history0.history['val_loss'], label='val_loss')
       plt.legend()
       plt.show()
       count=0
       for i in history0.history['acc']:
           if i>0.99:
               count+=1
       print(count)
   
  

       #Using 4 Sigmoid Layers and RMSprop optimizer
       model4 = Sequential()
       model4.add(Dense(500, activation='sigmoid', use_bias=True, input_shape=(30,)))
       model4.add(Dense(500, activation='sigmoid', use_bias=True))
       model4.add(Dense(500, activation='sigmoid', use_bias=True))
       model4.add(Dense(1, activation='sigmoid'))
       keras.optimizers.RMSprop(lr=0.01, rho=0.9, epsilon=None, decay=0.0)
       model4.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['accuracy'])
       history1 = model4.fit(mat[:,1:],mat[:,0], validation_split=0.3,shuffle=False,epochs=3000, batch_size=128, verbose=0)
       #model.evaluate(mat[:,1:],mat[:,0], batch_size=None, verbose=1, sample_weight=None, steps=None)
       score = model4.evaluate(mat[:398,1:],mat[:398,0], verbose=0, batch_size=128)
       print('Train loss:', score[0])
       print('Train accuracy:', score[1])
   
  

       Train loss: 0.1454299822000403
       Train accuracy: 0.9346733668341709
   
  

       score = model4.evaluate(mat[398:,1:],mat[398:,0], verbose=0, batch_size=128)
       print('Validation loss:', score[0])
       print('Validation accuracy:', score[1])
   
  

       Validation loss: 0.33671002412400053
       Validation accuracy: 0.8830409339296887
   
  

       plt.plot(history1.history['acc'], label='acc')
       plt.plot(history1.history['val_acc'], label='val_acc')
       plt.legend()
       plt.show()
       plt.plot(history1.history['loss'], label='loss')
       plt.plot(history1.history['val_loss'], label='val_loss')
       plt.legend()
       plt.show()
       count=0
       for i in history1.history['acc']:
           if i>0.99:
               count+=1
       print(count)
   
  

Connecting artificial intelligence (AI) with pharmaceutical sciences.

参考资料：

https://github.com/vkasojhaa/Clinical-Decision-Support-using-Machine-Learning