Accurate and efficient energy management in residential and commercial buildings can be achieved by providing real-time monitoring of appliances and providing detailed insight to consumers about their consumption behaviors. These insights identify energy wastage and support demand-side management. Non-intrusive load monitoring provides disaggregated energy information of target appliances by observing the feature variations in aggregate demand. Nowadays, most residential appliances are non-linear and draw non-sinusoidal currents, therefore employing fundamental active and reactive power as input features results in degraded performance of load identification algorithms. To deal with this drawback, this research proposes an improved load identification method which incorporates the fundamental and harmonic characteristics of current and voltage besides the active power and reactive power features. In this work, 21 features computed by IEEE standard 1459, harmonic energy distribution, wavelet transformation, spectral flatness measures, etc. are analyzed for appliance recognition. To select the prominent features, regularized neighborhood component analysis (NCA) is applied. The selected features are then applied to boosted tree classifier (BTC). The proposed load identification approach is validated using high-frequency start-up events of appliances from WHITED and COOLL datasets. Results obtained by the proposed technique surpass various recent techniques in classifying various appliance activations.