The Arithmetic Residuals in K-groups Analysis (ARKA) framework has recently been proposed as a supervised dimensionality reduction technique, which has been shown to identify the activity cliffs. In this approach, the training data points are partitioned into active and inactive classes; the descriptors contributing more to each class are identified using the most discriminating features approach. Finally, the weightage of each descriptor for the respective classes is computed, and the entire descriptor matrix is reduced to two dimensions: ARKA_1 and ARKA_2. The scatter plots of the data points in the reduced dimension space can identify the activity cliffs. The descriptors ARKA_1 and ARKA_2 have also been shown to model binary classification of small data sets. More recently, a multi-class ARKA framework has been proposed, considering the contribution of different QSAR descriptors to different activity ranges. This approach has been useful in regression-based model development, particularly when integrated with the similarity-based quantitative read-across structure-activity relationship (q-RASAR) modeling.