Segmenting Peruvian Firms by Innovation Patterns: A Hybrid Approach Using Clustering and Neural Networks

This paper proposes a hybrid analytical framework integrating unsupervised and supervised machine learning techniques to segment Peruvian firms according to their innovation profiles. Drawing upon microdata from the 2018 National Innovation Survey, two complementary sets of vari-ables-binary indicators of innovation activities (block C3P1) and continuous measures of associated investment (block C3P1_E)-were combined to capture both behavioral and financial dimensions of innovation. Dimensionality reduction was performed using Principal Component Analysis (PCA) to address high dimensionality and multicollinearity. Subsequently, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) was applied to identify natural groupings, including outlier cases. To validate and operationalize the segmentation, a Multilayer Perceptron neural network was trained to predict cluster membership, achieving an accuracy of 99%. The analysis revealed three predominant innovation profiles, differentiated by activity scope, investment levels, and structural characteristics. The proposed methodology offers a replicable, data-driven approach for strategic segmentation, with potential applications in innovation policy design, targeted support programs, and real-time firm diagnostics.