A Stochastic Optimization-Based Artificial Neural Fuzzy Inference Method for Financial Time Series

Abstract

This invention relates to a stochastic optimization-based artificial neural fuzzy inference method for financial time series that can be used in financial time series analysis, market trend analysis, investment decision support systems, algorithmic trading systems, portfolio management systems, financial data analysis systems, financial technology applications, artificial intelligence-based forecasting systems, computational intelligence applications, data science applications, economic data analysis systems, time series forecasting systems, risk analysis systems, and financial modeling applications. Its features include a processor (2) that runs the system components, a data collection and preparation module (3) that collects historical price and trading volume data from financial data sources and performs missing data correction, outlier removal, and data normalization operations, and a feature extraction module (4) that generates model input variables by creating RSI, MACD, and moving average technical indicators from the said data.It includes the module for establishing the ANFIS model, which performs reference model training with classical learning algorithms (5), the module for developing a stochastic optimization algorithm that generates stochastic-based optimization parameters for optimizing the membership function parameters of the ANFIS model (6), the module for integrating the algorithm that applies the generated optimization parameters to the membership functions of the ANFIS model into the ANFIS model (7), the module for model training and testing process, which performs the training and testing of the hybrid M-ANFIS model by creating a training dataset and a test dataset (8), and the module for performance analysis, which evaluates the results of the hybrid M-ANFIS model and the reference ANFIS model through MSE, RMSE and directional prediction accuracy metrics (9).

Description

1 TARIFF Stochastic optimization for financial time series. ARTIFICIAL NEURAL FUZZY INFERENCE METHOD BASED ON SYNTHETIC NEURAL FUZZY. Technological Field: This invention is used in financial time series analysis, market trend analysis, and investment decision support. systems, algorithmic trading systems, portfolio management systems, financial data analysis systems, financial technology applications, artificial intelligence-based prediction systems, 10 computational intelligence applications, data science applications, economic data analysis systems, time series forecasting systems, risk analysis systems, and financial modeling. Stochastic optimization for financial time series that can be used in applications. It is related to artificial neural fuzzy inference based on a specific method. State of the Art: Various statistical and artificial intelligence methods have been used in the literature for forecasting financial time series. Modeling based on and hybrid modeling methods is used. These methods include: In particular, Autoregressive Integrated Moving Average (ARIMA) models and artificial neural networks 20 (ANN), support vector regression (SVR), deep learning-based LSTM models, and Adaptive Neural Fuzzy Inference System (ANFIS) based approaches are widely used. is preferred. These methods utilize historical financial data. predicting future price movements, market trends, and investment behavior. It aims to do so. 25 ARIMA and similar econometric models used in known techniques, over time It involves mathematically modeling the statistical relationships in series, and especially Artificial neural networks can yield effective results in datasets with a linear structure. Networks and deep learning-based methods, on the other hand, reveal 30 differences between multidimensional datasets. It generates predictions by learning complex relationships. ANFIS systems, on the other hand, use fuzzy logic. by combining the advantages of artificial neural networks, it can analyze data containing uncertainty. 2 It enables the processing and modeling of nonlinear relationships. Learning algorithms for membership functions and inference rules in systems forecasting processes are optimized and the resulting parameters are used. is being carried out. However, financial markets are highly volatile and nonlinear. It produces data structures that are dynamic and contain a lot of noise. This occurs in the markets. sudden price movements, speculative transactions, crisis periods, and unusual trading volumes, This can significantly affect the performance of existing prediction models, especially. Traditional econometric methods and some machine learning algorithms can be used to analyze financial data. to represent complex patterns in data and variable market conditions with sufficient accuracy He / She is having difficulty doing so. Derivative or gradient-based models are commonly used in standard ANFIS models. Learning algorithms, optimization process on complex error surfaces 15 During implementation, it can get stuck at local minimum points. This situation affects the model. parameters in lower-performing solutions instead of global optimum solutions This leads to delays and limits the accuracy of predictions, especially with big data. This problem occurs in applications involving sets and the use of numerous input variables. is becoming more apparent, the generalization ability of the model and different markets 20 The stability of these conditions is negatively affected. In the literature, particle swarms are used to improve the performance of ANFIS models. optimization (PSO), genetic algorithms (GA) and similar meta-heuristic methods Although hybrid solutions have been proposed, a significant portion of these studies involve financial constraints. Reducing the noise effect in the data and local minimization in the optimization process. There is no integrated approach to addressing the problem together. Furthermore, transaction volume data is systematically used as a noise filtering mechanism. a comprehensive approach where it is used and integrated with the stochastic optimization approach The ANFIS architecture is not encountered. Therefore, financial time series are no longer available. The technical need for reliable modeling continues. 3 Description of the invention: This discovery represents an advancement in existing methods used for forecasting financial time series. to reduce the optimization deficiencies and noise sensitivity problems encountered It offers a solution for this purpose. The developed hybrid M-ANFIS (Modified Adaptive 5 Neuro-Fuzzy Inference System Thanks to the system-based structure, model parameters are optimized more effectively. This makes it possible to handle the complex nature of financial data more successfully. This allows for modeling. This is especially true for high volatility areas. It contributes to improving forecasting performance in the markets. 10 The stochastic-based hybrid optimization approach used in this invention is a standard one. It reduces the effect of the local minimum problem seen in ANFIS models. Thus, more suitable parameter combinations are obtained in the model learning process. being able to achieve and the optimization process being carried out more stably 15 This is possible. This feature allows the prediction system to perform better on different datasets. It enables the production of reliable results. The invention involves integrating transaction volume data and various technical indicators into the model. Thanks to this, random fluctuations and noise effects found in financial data are reduced by 20%. This is achieved by reducing the negative impact of indicators such as RSI, MACD, and moving averages. The additional information obtained allows the model to provide a more comprehensive understanding of market behavior. It helps in the evaluation and increases the accuracy of the prediction results. The developed hybrid structure offers 25 advantages compared to traditional approaches that rely solely on price data. They can evaluate more sources of information and thus better understand market trends. This allows for accurate analysis, especially during periods of crisis, boom, and bust. Maintaining the model's performance under different market conditions, such as those mentioned above, is crucial for the system. It expands the field of application. Another advantage of the invention is that the model's convergence performance has been improved. The model is made possible by integrating the stochastic optimization approach into the learning process. 4 The parameters can reach the appropriate values ​​in a shorter time, and the calculation processes... Efficiency can be increased. Thus, it is performed on large datasets. More efficient work can be achieved in the analyses. The developed method includes: stock price prediction, market trend analysis, and investment decision-making. support systems, portfolio management applications, and algorithmic trading systems It offers a flexible infrastructure that can be used. In addition, financial time series energy prices, economic indicators, and other time series-based forecasts. It provides a general-purpose modeling approach that can also be applied to these problems. The invention also offers 10% access to existing financial technology infrastructures thanks to its software-based structure. It can be easily integrated and provides data analytics and AI-based decision support. It can be used in their systems. This feature is used by financial institutions, brokerage firms, Commercial feasibility for portfolio management companies and individual investors increasing and having wide usage potential in national and international markets. It constitutes. 15 Explaining the Figures: The invention will be described by referring to the attached figures, so that the features of the invention can be explained. will be understood and appreciated more clearly, but the purpose of this invention is this particular 20 It is not about limiting it with regulations. On the contrary, the invention is defined by the accompanying claims. all alternatives, modifications, and options that could be included within the defined area The aim is to cover their equivalences. The details shown are only for the present invention. It is shown to illustrate the preferred arrangements and both the methods the shaping of both the rules and conceptual features of the invention in the most useful 25 It should be understood that they are presented to provide a readily understandable definition. This in the drawings; Figure 1 shows a schematic view of the system. Illustrations that will help understand this invention are shown in the attached image. They are numbered and their names are given below. Explanation of References: 1. Server 2. Processor 5 3. Data Collection and Preparation Module 4. Feature Extraction Module 5. Module for Establishing the Reference ANFIS Model 6. Development of Stochastic Optimization Algorithm Module Module 7: Integration of the Algorithm into the ANFIS Model (Module 10) 8. Model Training and Testing Process Module 9. Performance Analysis Module Description of the Invention: The invention includes a processor (2) that operates the system components, from financial data sources. It collects historical price and trading volume data, performs missing data correction, and detects outliers. Data collection and preparation that performs cleaning and data normalization processes. module (3), RSI, MACD and moving average technical data from the said data Feature extraction module 20 that generates model input variables by creating indicators. (4), ANFIS which performs reference model training with classical learning algorithms. the model setup module (5), the membership function of the ANFIS model Stochastic-based optimization parameters for the optimization of parameters the development module of the generated stochastic optimization algorithm (6), 25 that apply optimization parameters to the membership functions of the ANFIS model integration module of the algorithm into the ANFIS model (7), training dataset and test Training and testing of the hybrid M-ANFIS model by creating a dataset. the model training and testing process module (8) and with the hybrid M-ANFIS model The results for the reference ANFIS model include MSE, RMSE, and directional prediction accuracy. It includes the performance analysis module (9) which evaluates through metrics. 30 6 The invention is a data generator that collects historical price and trading volume data from financial data sources. It has a collection and preparation module (3). The invention corrects missing data, discrepancies Data collection and preparation that cleans values ​​and performs data normalization. It has module (3). The invention calculates RSI, MACD and moving average indicators. It has a feature extraction module (4). 5 The invention is a performance device that performs MSE, RMSE, and directional estimation accuracy calculations. It has an analysis module (9). The invention generates stochastic-based optimization parameters for stochastic optimization 10. It has a module (6) for developing the algorithm. The invention has a membership function premise. module for integrating the algorithm that optimizes its parameters into the ANFIS model (7) has. The invention creates a training dataset and a test dataset and uses a hybrid M-ANFIS model with 15 Model training and testing that performs comparative analysis of the reference ANFIS model. It has process module (8). The invention is an artificial neural fuzzy system based on stochastic optimization for financial time series. It is an inference method that involves collecting historical price and trading volume data, and missing data 20 correction, outlier removal, and data normalization processes the implementation of RSI, MACD and moving average technical indicators creation, establishment of the reference ANFIS model, stochastic optimization the development of the algorithm, the developed optimization algorithm ANFIS integration into the model, training and testing of the hybrid M-ANFIS model 25 and includes conducting performance analysis. The invention utilizes MSE, RMSE, and directional prediction in the performance analysis step. It calculates accuracy. The invention, in the step of creating technical indicators, uses RSI. It calculates MACD and moving average indicators. 30 7 The invention lies in the optimization step of developing a stochastic optimization algorithm. The parameters are being created. The invention is the optimization algorithm of ANFIS. In the step of integrating it into the model, the membership function's antecedent parameters are optimized. The invention involves training and testing a hybrid M-ANFIS model. Step 5 involves creating the training dataset and the test dataset. Detailed Description of the Invention: The invention is an artificial intelligence-based stochastic optimization tool for forecasting financial time series. Neural fuzzy inference system and hybrid modeling performed by this system 10 It is related to the method. The system in question involves the collection and processing of financial data. creation of technical indicators, establishment of the ANFIS model, stochastic optimization development of the algorithm, integration of the developed optimization algorithm into the ANFIS model integration, training, testing and performance analysis of the model It carries out its operations as a whole. 15 Within the scope of the invention, the system’s basic processing unit is located on the server (1). The processor (2) is the data collection and preparation module in the system. (3), feature extraction module (4), ANFIS model setup module (5), stochastic Development module of optimization algorithm (6), 20 of the algorithm to ANFIS model integration module (7), model training and testing process module (8) and performance analysis It manages the operation of the module (9). In this context, the processor (2) manages the operation of the modules. providing data flow and calculations related to the financial time series forecasting process. It carries out its operations. When the system starts, the data collection and preparation module (3), financial data It collects historical price and trading volume data from various sources. Data collection and preparation module (3), missing data on collected price and trading volume data It performs the correction process. The same module (3) finds the outlier in the dataset. 30 in cleaning values ​​and modeling financial time series data in the process The data collection and preparation module (3) brings it into a usable format. 8 normalizing the dataset after processing and extracting features from the processed dataset. It transfers to module (4). Feature extraction module (4) normalizes data from the data collection and preparation module (3). Using collected financial time series data, the model input variables are 5 It creates. The feature extraction module (4) adds raw price and trading volume data. It calculates the RSI indicator. The same feature extraction module (4) calculates the MACD. It calculates the indicator and creates moving average indicators. This The technical indicators generated in this way are organized as the input variables of the model. and is transferred to the ANFIS model setup module (5). 10 ANFIS model setup module (5), input from feature extraction module (4) It constructs the reference ANFIS model using its variables. The ANFIS model establishment module (5), reference model training with classical learning algorithms This process is carried out according to rule 15 of the ANFIS structure. The base and inference structure are being created. ANFIS model establishment module (5), It generates initial estimation results based on the reference model and uses these results make available within the system for subsequent comparative analysis processes It brings. Development module of the stochastic optimization algorithm (6), belonging to the ANFIS model Stochastic basis to be used in the optimization of membership function parameters It generates the optimization parameters. The module in question (6) is the ANFIS model. Optimization starting values ​​for use in the learning process, search It creates the parameters and parameter update values. Stochastic 25 Development module of optimization algorithm (6), membership function It describes stochastic search operations for updating parameters and parameter sets to be used in the optimization process are entered into the algorithm's ANFIS model. It transfers to the integration module (7). Integration module of the algorithm into the ANFIS model (7), stochastic optimization optimization parameters from the algorithm development module (6) 9 It applies the ANFIS model to membership functions. The algorithm is based on ANFIS. Integration module into the model (7), membership function precursor of the ANFIS model It optimizes its parameters. This module (7) uses a stochastic optimization algorithm. incorporating the parameter values ​​generated by the system into the learning process of the ANFIS model, and It enables the creation of a hybrid M-ANFIS model. Thus, the reference ANFIS 5 The model is a hybrid M-ANFIS that works in conjunction with a stochastic optimization algorithm. It is being converted into a model. Model training and testing process module (8), training and testing of the hybrid M-ANFIS model It performs its operations. Model training and testing process module (8), data set 10 It separates the training dataset and the test dataset. The module in question (8), Learning process of hybrid M-ANFIS model using training dataset It is being carried out. After the training process is completed, the model training and testing process module (8), prediction results of the hybrid M-ANFIS model from the test dataset It produces the same module (8), the hybrid M-ANFIS model and the reference ANFIS model 15 It is preparing its results for comparative analysis. Performance analysis module (9), from model training and testing process module (8) It evaluates the prediction results. Performance analysis module (9), hybrid M- The results of the ANFIS model and the reference ANFIS model are compared using the MSE error metric. It calculates the RMSE error metric. The same performance analysis module (9) calculates the RMSE error metric. It performs calculations and direction prediction accuracy calculations. The performance analysis module (9) uses these metrics to develop a hybrid M-ANFIS model. This completes the comparative performance analysis of the reference ANFIS model. 30

Claims

REQUESTS 1- The invention is an artificial neural network based on stochastic optimization for financial time series. It relates to the fuzzy inference system, and its characteristic feature is;  a server (1) and a processor (2) that runs the system components, 5  collects historical price and trading volume data from financial data sources, missing data correction, outlier removal, and data normalization processes data collection and preparation module (3),  from the data in question, RSI, MACD and moving average technical indicators feature extraction module (4), 10 that generates model input variables by creating  ANFIS performs reference model training using classical learning algorithms. model establishment module (5),  Optimization of the membership function parameters of the ANFIS model Stochastics that generate stochastic-based optimization parameters for Development of optimization algorithm module (6), 15  The generated optimization parameters are added to the membership functions of the ANFIS model. Integration module of the algorithm implementing into the ANFIS model (7),  Creating a training dataset and a test dataset to develop a hybrid M-ANFIS model Model training and testing process module (8), which performs training and testing operations,  Results of the hybrid M-ANFIS model and the reference ANFIS model MSE, RMSE 20 and performance analysis evaluated through direction prediction accuracy metrics. It consists of module (9). 2- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system, characterized by its use of historical prices and 25% of financial data sources. It has a data collection and preparation module (3) that collects transaction volume data. It is the characterization of the situation. 3- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system whose characteristic is that it corrects missing data and removes outliers. data collection and preparation module that cleans and performs data normalization (3) is characterized by having. 11 4- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system, characterized by its use of RSI, MACD, and moving average indicators. It is characterized by having a feature subtraction module (4) that calculates. 5- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system, characterized by its accuracy in MSE, RMSE, and directional prediction. with having a performance analysis module (9) that performs the calculations It is the characterization of the situation. 6- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system, characterized by its stochastic-based optimization parameters. having a module (6) for developing the generated stochastic optimization algorithm It is characterized by... 7- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system, characterized by its ability to optimize the antecedent parameters of the membership function. with the integration module (7) of the algorithm into the ANFIS model It is the characterization of the situation. 8- Claim 1: Stochastic optimization-based artificial neural network for financial time series. It is a fuzzy inference system, characterized by the creation of a training dataset and a test dataset. and a comparative analysis of the hybrid M-ANFIS model and the reference ANFIS model. It is characterized by having a model training and testing process module (8) which carries out the model training and testing process. It is done. 25 9- The invention is an artificial neural network based on stochastic optimization for financial time series. It is a fuzzy inference method, and its characteristic is;  Collection of historical price and trading volume data,  30 of the processes of correcting missing data, removing outliers and normalizing data. to be carried out,  Creating technical indicators such as RSI, MACD, and moving averages, 12  Establishment of the reference ANFIS model,  Development of a stochastic optimization algorithm,  Integration of the developed optimization algorithm into the ANFIS model,  Training and testing of the hybrid M-ANFIS model,  Performing a performance analysis involves the following steps: 5 10- Stochastic optimization based on financial time series mentioned in claim 9. It is an artificial neural fuzzy inference method, and its characteristic feature is performance analysis. In the implementation step, MSE, RMSE and directional forecast accuracy calculations are performed. It is characterized by the fact that it involves being done. 10 11- Stochastic optimization based on financial time series mentioned in claim 9. It is an artificial neural fuzzy inference method, and its characteristic feature is that it uses technical indicators. Calculation of RSI, MACD and moving average indicators in the creation step It is characterized by... 15 12- Stochastic optimization based on financial time series mentioned in claim 9. It is an artificial neural fuzzy inference method, characterized by its stochastic optimization properties. in the algorithm development step by creating optimization parameters It is characterized. 20 13- Stochastic optimization based on financial time series mentioned in claim 9. It is an artificial neural fuzzy inference method, and its characteristic feature is that the optimization algorithm... In the step of integrating the ANFIS model, the membership function prerequisite parameters It is characterized by its optimization. 25 14- Stochastic optimization based on financial time series mentioned in claim 9. It is an artificial neural fuzzy inference method, and its characteristic feature is the hybrid M-ANFIS model. In the training and testing phase, the training dataset and the test dataset It is characterized by its creation. 30

Images