Abstarct: Nowadays, the use of neural networks in line with predictions that have higher reliability is very important. Neural networks are part of intelligent systems that, by processing experimental data, transfer the knowledge or law behind the data to the network structure, hence they are called intelligent systems. Stock market price forecasting is also a hot topic in the research and application of neural networks. An important step in the application of a neural network is its design. Generally, the data in the companies' baxse databaxses or databaxses are selected and corrected to create a suitable data set for the design. Then the choice of neural network architecture and design of input data are important issues of predictive neural networks. In this study, an unconventional method for stock forecasting is presented that uses a Recurrent Neural Network to determine "buy", "sell" and "hold" scenarios through images made from stock charts. In this model, time series data is converted into a set of images consisting of stock price bar charts. Each converted image contains stock price and time. Also, each image contains information representing 30 days of stock price. Our goal in this study is to estimate and present a suitable model for predicting the price of raw stock. In this regard, we first modeled the stock price series structure baxsed on a nonlinear model baxsed on neural network and then by applying this method to decision tree and random forest algorithms, we performed a species comparison for them. The results indicate that the model implemented in the design, due to the nature of the option under consideration (stock exchange) has a significant performance in accuracy compared to other models and the baxse model. Finally, the proposed method provides 90.4% classification accuracy for diagnosing stock status.