This paper presents implementation of new features for classification of bearing faults in diagnosing induction motor. The proposed features are time dependent spectral features, a set of descriptors of frequency domain features are directly derived from time-domain vibration data of present and previous window segment. The higher order moments and hence logarithmic features are computed for a segment and from nonlinear scaling of same segment to obtain descriptors of the window. The orientation is computed from the descriptors of window segment and non-linear window segment. The orientation of features of current window and previous window is combined to obtain fusion features. The fused features are utilized for three case study data, derived from publicly available database of Case Western Reserve University. The performance of fusion features are compared with conventional features for 10-fold cross validation using linear discriminant analysis and naive Bayes classifiers. The results have shown, the proposed features improve the accuracy of classification and with decrease in computational cost of12.2%, compared to conventional features. In addition, average of 2% increase in accuracy of classification is obtained, compared other of classification. © 2018, Institute of Advanced Scientific Research, Inc. All rights reserved.