2. PROBLEM STATEMENT
The objective of project was to classify the students
into two categories ‘Male’ and ‘Female’.
This classification is to be done on the basis of their
educational record.
We have to approach the classification by applying
unsupervised learning approach i.e kmeans and
any supervised learning approach.
3. SOLUTION APPROACH
Data Collection and Data entrance:
We collected the data of students targeting some
attributes which can be useful In decision making .
Data load in Matlab:
The data files were loaded in to Matlab
Application of Kmeans:
Matlab queries were applied on the data to get the
result
Data Division:
The whole data was divided into 70% of male and
female for Train dataset. Remaining 30% was
divided into Male and Female Test dataset.
Knn Application ; knn was applied on test and train
data
4. DATA SET
Rows 174
Features 8 (Category, sgp1, sp2, cgp, total
marks, database gp , oodm and
graphics gp)
Female data 111 rows
Mal data 63 rows
9. The Blue shaded area left to the 0 of cluster1 along
x-axis is misclassified because it is towards
negative value
Cluster2 is better classified than cluster1 because it
is more towards positive 1 value.
RESULT OF KMEANS IN THE FORM OF SILHOUTTE
GRAPH
10. KNN ALGORITHM
Traning data:
70% training data of females=77 rows
70% trainning data of males=44
Testing data:
30% testing data of females=34
30% testing data of males=19
11. RESULT OF KNN WITH DIFFERENT K
Value of k %Training accuracy %Testing
accuracy
1 98.3871 50
2 71.7742 56
3 70.1613 52
5 67.7419 50
7 65.3226 52
8 60.4839 58
9 63.7097 54
12. PLOT OF TRAIN AND TEST ACCURACIES WITH
DIFFERENT K
0
20
40
60
80
100
120
k1 k2 k3 k5 k7 k8 k9
train acc
test accu
13. ANOTHER GRAPH OF RESULT OF KNN
0
10
20
30
40
50
60
70
80
90
100
k1 k2 k3 k5 k7 k8 k9
test accu
train accu
15. COMPARISON OF RESULTS OF ALL
ALGORITHMS
Algorithm Trainning
accuracy
Testing
accuracy
accuracy error
kmeans --------- ----------- 58.7 41.3
Knn with
k=2
71.7742 56 ---------- -----------
Ann with
h=10
65.6 73.1 ----------- ------------
16. CONCLUSION
Accuracy of unsupervised algo (kmeans) was
improved by supervised learning algorithms knn
and ann.
Training accuracy of knn is better than Ann at value
k=2.
Testing accuracy of Ann is greater then knn at
hidden layers 10.