MLmodel_skin.py

import os
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.svm import SVC
import pickle
import numpy as np
from skimage.io import imread
from skimage.transform import resize

"""
This file is used to load the data and train the model for detecting melanoma.
"""

#load and process data
path = os.listdir("./train_skin")
id = {'benign':0,'malignant':1}
images, ids = [], []
for folder in id:
    path2 = "./train_skin" + "/" + folder
    for item in os.listdir(path2):
        img = imread((path2 + '/' + item), as_gray=True)
        img_resized = resize(img, (200,200))
        img_resized = img_resized.reshape(1, -1) / 255.0
        images.append(img_resized)
        ids.append(id[folder])
images = np.array(images)
ids = np.array(ids)
images = images.reshape(len(images), -1)
x_train, x_test, y_train, y_test = train_test_split(images, ids, random_state=10,
                                                    test_size=.2)

#feature scaling
x_train = x_train / 255
x_test = x_test / 255

#train
svc = SVC()
svc.fit(x_train,y_train)

#evaluate
print("Training score:", svc.score(x_train, y_train))
print("Testing Score:", svc.score(x_test, y_test))

#test
decode = {0 : 'benign', 1 : "malignant"}
prediction, correct= [], []
for i in os.listdir("./test_skin/benign/"):
    img = imread(("./test_skin/benign/" +i), as_gray=True)
    img_resized = resize(img, (200,200))
    img_resized = img_resized.reshape(1, -1) / 255.0
    p = svc.predict(img_resized)
    prediction.append(p[0])
    correct.append(0)

for i in os.listdir('./test_skin/malignant/'):
    img = imread(('./test_skin/malignant/'+i), as_gray=True)
    img_resized = resize(img, (200,200))
    img_resized = img_resized.reshape(1, -1) / 255.0
    p = svc.predict(img_resized)
    prediction.append(p[0])
    correct.append(1)

print("accuracy score:", accuracy_score(prediction, correct))

#uncomment to save modek as a pkl file
#pickle.dump(svc, open('detect_melanoma.pkl','wb'))