Michitaka Kosaka

Applications of fuzzy logic/neural network to securities trading decision support system

The effectiveness of applying fuzzy logic and neural networks to securities trading decision support systems (STDSS) is demonstrated through some examples. First, the characteristics of STDSS are discussed. Next, examples such as buy/sell timing detection or stock portfolio selection using fuzzy logic and neural networks are reported by showing their algorithms and simulation results. In a test of the models ability to follow price trends, the model correctly identified 65% of all price turning points.<<ETX>>

A track correlation algorithm for multi-sensor integration

A track correlation algorithm in a multi-sensor integration system for a surveillance mission is proposed. The performance of such track correlation processing is strongly dependent not only on the target state estimation error distribution but also on the target spatial density distribution. Therefore, a track correlation problem is formulated as the likelihood ratio test problem which can take both target state estimation error distribution and target state spatial density distribution into consideration. From this formulation, the correlation algorithm for on-line processing is derived by modifications and approximations. Through analytical evaluations and simulation studies, it is shown that the proposed algorithm is superior to the conventional nearest neighbor algorithm.

Evaluation Method of Polynomial Models' Prediction Performance for Random Clock Error

In satellite navigation systems such as the Global Positioning System, clock error is one of the major sources of error in precise pointing. In order to remove clock error, it is modeled as a second-order polynomial and the clock-error correction parameters are sent to users. However, a random clock error cannot be modeled as a second-order polynomial. Therefore, the time discrepancies due to random clock error must be taken into consideration for precise pointing. This paper proposes an analytical computation method for estimating the random clock error in the current system which makes use of the Allan variance characteristics of random clock error without random clock realization and a lot of simulation studies. Moreover, a numerical example based on the proposed method shows that the first-order polynomial model is better for predicting a random clock error than the second-order polynomial.

Addendum to "Smoothing Algorithms for Two-Dimensional Image Processing"

It should be noted that the state-space model used in the above correspondence is not a Markov model for homogeneous image fields and that the algorithms derived are not optimal in any sense. Therefore, the algorithms should be understood as suboptimal.

A Decision Support System Building Tool with Fuzzy Logic and Its Application to Chart Technical Analysis

Abstract Financial decision support systems (DSS), including securities trading support systems or asset liability management systems, have become more intelligent and sophisticated due to the progress of information processing technology, particularly in areas of artificial intelligence (AI). Knowledge acquisition often bottlenecks during the building of intelligent DSS (IDSS) based on AI techniques. This paper describes a DSS building tool with fuzzy logic builtin, which supports knowledge acquisition functions using human-machine interaction and various simulation functions. This building tool consists of the following functions; (1) Fuzzy reasoning, (2) User-friendly fuzzy rule editor, (3) Time-series data handling, including chart graphics, (4) and Simulation and rule evaluation. We applied this tool to build a chart technical analysis support system. This system stores technical analysis know-how in the fuzzy knowledge base, provides various types of information, and helps decision makers make more informed decisions and acquire new knowledge. The effectiveness of knowledge acquisition using this tool has been demonstrated through its application.

A Method of Supporting Price Chart Analysis Using Flexible Template Matching

Abstract Price chart analysis is a powerful technique of market analysis. Human experts interpret chart and extract characteristic patterns related to subsequent price changes. For computer systems to have this ability, pattern matching needs to incorporate experts know-how. This paper proposes a method of chart analysis using template matching, applicable to Point & Figure. In the method, templates of the characteristic patterns are represented so their shapes can vary under constraints corresponding to the know-how. By comparing with the templates, characteristic patterns including variations are extracted. Then, buying-and-selling signals are generated. A simulation using practical data demonstrates effectiveness of the method.

Recursive least square filtering for two-dimensional radom fields

An approximate two-dimensional recursive filtering algorithm that parallels Kalman filter is presented for a causal system considered in [1].