Constructing a portfolio from a large number of active stocks is a critical as well as challenging investment decision due to high volatility and biased decision making. The abundance and availability of _nancial data gives machine learning (ML) an advantage to optimize investment decisions. The k-means algorithm is used to cluster observations into di_erent groups, each of which contains those with similar properties. In this paper, three factors are considered to cluster stocks and select clusters with best performing stocks for portfolio construction. It enhances the cardinal investment decision of stock selection to construct optimized portfolios. The out-of-sample performance demonstrates high economic gains from the proposed strategy with an average Sharpe ratio of 0.7.