With the considerable amount of data collected from the Information Technology systems including computer networks, many institutions and organizations are considering the benefit in discovering user patterns. This paper presents the processes in developing the enhanced algorithm for Linear Discriminant Analysis via Generalized Singular Value Decomposition (LDA/GSVD) which primarily functions to discover web usage patterns. Preprocessing, class labeling using Self-Organizing Map (SOM), computation of the feature subspaces, and the training of the Artificial Neural Network (ANN) architecture are stages in developing the algorithm. After the development, this neural network-based algorithm was used to discover patterns from the university’s proxy servers’ weblogs. Results showed an appropriate classification of the network users’ web usage with the improved algorithm. Hence, results revealed that more users in the campus whether students or employees are utilizing the internet for noneducational rather than educational. Also, simulation results showed that the enhanced algorithm outperformed the current LDA/GSVD algorithm up to 50% on computational cost. Through this approach, IT managers will be guided to make better plans to optimize the utilization of the internet. Moreover, this proposed technique will benefit not only the educational institutions but also other organizations which mostly need pattern discovery to their systems’ enormous data.