Construction of optimal portfolios is one of the concerns for financial institutions, public funds and individual investors. Generally, in portfolio optimization problem we consider at least two conflicting objective functions – minimizing portfolio risk and maximizing portfolio reward. Additionally, each investor or fund manager has their own preferences, therefore we typically add the set of constraints. A resulting optimization problem is typically a high-dimensional, computationally challenging nonlinear optimization problem that may require developing specialized formulations and novel solution techniques. In this research project, we plan to investigate and test three strategies for building sparse portfolios (where we limit a number of assets) and compare performances with baseline portfolios. First strategy is cardinality-constraint optimization. Second strategy is regularized optimization approximation. Third strategy is to utilize Machine Learning clustering algorithms to group similar assets and select a most representative one from each cluster prior to using those assets as a trading universe in optimization. Partner organization plans to enhance its software products with advanced sparse portfolio optimization formulations.