-> The MNIST dataset is an acronym that stands for the Modified National Institute of Standards and Technology dataset. (for more info visit their link:- http://yann.lecun.com/exdb/mnist/ )
-> It is a dataset of 60,000 small square 28×28 pixel grayscale images of hand written single digits between 0 and 9.
-> The task is to classify a given image of a handwritten digit into one of 10 classes representing integer values from 0 to 9, inclusively. Ans find precesion matrix values such as Precision, Recall, F1 score, Sensitivity, Specificity and Accuracy.
The code is written in python language with necessary libraries and comments are there for every step to understand it better.
Result:-
Output for 5 times runing code for full data set.
Epoch 1/5
1875/1875 [==============================] - 33s 18ms/step - loss: 0.4723 - accuracy: 0.8328
Epoch 2/5
1875/1875 [==============================] - 33s 18ms/step - loss: 0.3178 - accuracy: 0.8871
Epoch 3/5
1875/1875 [==============================] - 33s 18ms/step - loss: 0.2767 - accuracy: 0.8997
Epoch 4/5
1875/1875 [==============================] - 33s 18ms/step - loss: 0.2503 - accuracy: 0.9089
Epoch 5/5
1875/1875 [==============================] - 33s 18ms/step - loss: 0.2314 - accuracy: 0.9155
predictoin and actual values for checking, is the model correct ?
predicted: [9 2 1 1 6 1 4 6 5 7 4 5 8 3 4 1 2 4 8 0 2 5 7 5 1 6 6 0 9 3]
Actual : [9 2 1 1 6 1 4 6 5 7 4 5 7 3 4 1 2 4 8 0 2 5 7 9 1 4 6 0 9 3]
Here, 100 % accurate for 30 digit-images out of 60,000 test images. Let's proceed on full testing data.....
Precision Matrix results are here that is created by using Convolutional Neural Networks (CNN) with sequential model adding relu and softmax activation function and adam optimizer to fit data.
for digits: 0 1 2 3 4 5 6 7 8 9
TP: [835 984 852 951 862 981 703 971 989 960] (True positive)
TN: [8895 8990 8848 8858 8832 8983 8786 8957 8969 8970] (True Negative)
FP: [105 10 152 142 168 17 214 43 31 30] (False Positive)
FN: [165 16 148 49 138 19 297 29 11 40] (False Negative)
Finally, results are...(0 to 9 stands for hand written digits)
precision specificity F1_scores accuracy recall or sensitivity
0 0.888298 0.988333 0.860825 0.9730 0.835
1 0.989940 0.998889 0.986961 0.9974 0.984
2 0.848606 0.983111 0.850299 0.9700 0.852
3 0.870082 0.984222 0.908743 0.9809 0.951
4 0.836893 0.981333 0.849261 0.9694 0.862
5 0.982966 0.998111 0.981982 0.9964 0.981
6 0.766630 0.976222 0.733438 0.9489 0.703
7 0.957594 0.995222 0.964250 0.9928 0.971
8 0.969608 0.996556 0.979208 0.9958 0.989
9 0.969697 0.996667 0.964824 0.9930 0.960
Thanks for your patience to understanding clearly.