Training model and Evaluation
Inception model evaluation.ipynb
IPython notebook for evaluating inception model
Training Classifier.ipynb
IPython notebook for training xgboost classifier
testing.csv
testing dataset
train_test.csv
training and testing dataset
training.csv
training dataset
xgb.model
classification model
Example of classification results
Image name
1st result
1st score
2nd result
2nd score
3rd result
3rd score
4th result
4th score
5th result
5th score
15.38203756895384-100.1636399994293-120-2013-09-5-Tree
lakeside, lakeshore
0.2218
worm fence
0.110338
swing
0.0501
golf ball
0.0215
golfcart, golf cart
0.0179
Evaluation of Inception Model (Image Recognition)
Performance of inception model (All classes)
named
Top-5
Top-1
Top-1 & thresh=0.15
Top-1 & thresh=0.30
sample size
60920
13274
7456
3739
accuracy
0.2774
0.3837
0.5053648
0.6060
Performance of inception model (Each class)
named
flowerpot
stupa, tope
water jug
water bottle
trash can
greenhouse
milk can
barrel, cask
canoe
rain barrel
lakeside
Dutch oven
Top-5_size
11589
232
648
208
5587
9850
3017
1071
1737
2008
25291
496
Top-5_acc
0.4220
0.2759
0.4012
0.4087
0.2980
0.3637
0.4475
0.5770
0.3172
0.5364
0.2355
0.4798
Top-1_size
4236
33
48
40
925
1552
470
109
182
319
5292
68
Top-1_acc
0.4498
0.48485
0.4583
0.475
0.4497
0.3937
0.5766
0.5596
0.5879
0.6677
0.2674
0.5441
Thresh_size
2862
17
21
19
476
961
229
67
95
206
2471
32
Thresh_acc
0.5346
0.7647
0.7143
0.5263
0.6492
0.4412
0.8165
0.7015
0.6526
0.7621
0.4027
0.5938
Evaluation of Machine Learning Classifier
Classification based on Top-5 results
Sample sizes
Yes
No
Ratio
Dataset
60920
16902
44018
0.28
Training set
50920
11567
38723
0.23
Testing set
10000
5000
5000
0.50
Classification Performance of XGBoost
Training
Testing
Accuracy
0.7666
0.5025
Classification based on Top-1 result
Sample sizes
Dataset
60920
Select where top-1 score > 0.15
32658
Select where top-1 = breeding sites
7456
Sample sizes
Yes
No
Ratio
Training set
5219
2657
2,562
0.5091
Testing set
2237
1111
1126
0.4966
Classification Performance
Individual classifier
Voting classifier
Final Classifier
SVM
Decision Tree
Knn
Random Forest
SVM + KNN + Decision Tree
Logistic Regression + Random Forest + GaussianNB
XGBoost
0.5739
0.5928
0.5928
0.6366
0.6165
0.6477
0.6822