In the complex and dynamic stock market landscape, investors seek to optimize returns while minimizing risks associated with price volatility. Various innovative approaches have been proposed to achieve high profits by considering historical trends and social factors. Despite advancements, accurately predicting market dynamics remains a persistent challenge. This study introduces a novel deep reinforcement learning (DRL) architecture to forecast stock market returns effectively. Unlike traditional approaches requiring manual feature engineering, the proposed model leverages convolutional neural networks (CNNs) to directly process daily stock prices and financial indicators. The model addresses overfitting and data scarcity issues during training by replacing conventional Q-tables with convolutional layers. The optimization process minimizes the sum of squared errors, enhancing prediction accuracy. Experimental evaluations demonstrate the model's robustness, achieving a 67% improvement in directional accuracy over the buy-and-hold strategy across short-term and long-term horizons. These findings underscore the model’s adaptability and effectiveness in navigating complex market environments, offering a significant advancement in financial forecasting.

All projects require a structure to meet project requirements and achieve established goals. This framework is called project management. Therefore, project management plays an important role in national development and economic growth. Project management includes various knowledge areas such as project integration management, project scope management, project schedule management, etc. The article focuses on the resource-constrained project scheduling known as problem so- called the resource-constrained project scheduling problem (RCPSP). The RCPSP is a part of schedule management. The standard RCPSP has two important constraints, resource constraints and precedence relationships of activities during project scheduling. The objective of the problem is to optimize and minimize the project duration, subject to the above constraints. In this paper, we develop a convolutional neural network approach to solve the standard single mode RCPSP. The advantage of this algorithm over conventional methods such as metaheuristics is that it does not need to generate many solutions or populations. In this paper, the serial schedule generation scheme (SSGS) is used to schedule the project activities using an evolved convolutional neural network (CNN) as a tool to select an appropriate priority rule to filter out a candidate activity. The evolved CNN learns according to the eight project parameters, namely network complexity, resource factor, resource strength, average work per activity, etc. The above parameters are the inputs of the network and are recalculated at each step of the project planning. Moreover, the developed network has priority rules which are the outputs of the developed neural network. Therefore, after the learning process, the network can automatically select an appropriate priority rule to filter an activity from the eligible activities. In this way, the algorithm is able to schedule all project activities according to the given project constraints. Finally, the performance of the Convolutional Neural Network (CNN) approach is investigated using standard benchmark problems from PSPLIB in comparison to the MLFNN approach and standard metaheuristics.