A Novel Risk-based Portfolio Optimization with Return Prediction Using Enhanced Heap Optimizer Aided Serial Cascaded Deep Residual Networks

The prediction of the stock market is still a complicated task to pursue in the field of machine learning as it relies on previous and past data, which is generally unstable and noisy. Existing methods of predicting the stock market are usually supervised approaches in which the prior labeling of the data as negative as well as positive moments is necessary to provide accurate results. However, this method of training the machine learning algorithms is complicated as it is vulnerable to overfitting issues as the behavior of the market usually relies on certain constraints like political events, marketing trends, and so on. The functioning of the actual portfolio optimization framework can be enhanced by fusing the return prediction in the conventional time series systems in the process of generating portfolios. Thus, the implementation of a better portfolio optimization system, which has a return prediction system integrated along with making the investors obtain more stable profits, is the major goal of this work. The modeling of the return forecasting model initially begins with the gathering of past stock marketing data. Then, the returns are estimated with the development of the Serial Cascaded Deep Residual Networks (SCDRN), where the deep learning models are utilized for predicting the future return of each stock. The risk level of each stock is also determined during the return prediction process. Then, with the aid of the developed optimization strategy named Enhanced Heap-based optimizer (EHO), the optimal portfolio is formulated, where the up-to-date market conditions are captured to rebalance the portfolio. The optimal selection of asset trading systems functions on the basis of the Sharpe ratio, and it maximizes the profit relative to the risk taken. An extensive analysis is executed to test the efficacy of the proposed portfolio optimization model regarding several baseline works with heuristic strategies.