NONSTATIONARY STREAM DATA LEARNING WITH IMBALANCED CLASS DISTRIBUTION - Imbalanced Learning: Foundations, Algorithms, and Applications - page 178

Table 7.3 Time and Space Complexity for SEA Dataset
Complexity

Normal SMOTE UB SERA REA

Training time
complexity (s)

2 . 632 4 . 092 7 . 668 3 . 952 5 . 356

Testing time
complexity (s)

0 . 088 0 . 148 0 . 236 0 . 156 1 . 904

Space complexity (kb) 1142 1185 1362 1266 1633

Table 7.4 Time and Space Complexity for ELEC Dataset
Complexity Normal SMOTE UB SERA REA

Training time
complexity (s)

1 . 376 1 . 712 2 . 844 1 . 824 2 . 028

Testing time
complexity (s)

0 . 084 0 . 148 0 . 244 0 . 14 1 . 744

Space complexity (kb) 129 142 155 151 248

Table 7.5 Time and Space Complexity for SHP Dataset
Complexity Normal SMOTE UB SERA REA

Training time
complexity (s)

10 . 29 17 . 89 60 . 10 16 . 98 21 . 62

Testing time
complexity (s)

0 . 23 0 . 46 0 . 53 0 . 43 11 . 58

Space complexity
(kb)

6881 7061 7329 9858 10,274

Considering the empirical performance of all algorithms on various datasets,
it could be concluded that SERA has the best performance-complexity trade-off.
That being said, as the consumption of time and space of REA is not significantly
larger than that of SERA, applications that do not place strong requirements on
speed and limitations on computational resources should as well consider using
REA because of its exceptional learning performance.

7.5 CONCLUSION

This chapter reviews algorithms for learning from nonstationary data streams with
imbalanced class distribution. All these algorithms basically manage to increase
the number of minority class data within the data chunk under consideration to
compensate the imbalanced class ratio, which can be categorized into the ones
using an over-sampling technique to create synthetic minority class instances,