Abstract

Technical chart patterns serve as crucial indicators for algorithmic trading systems, yet automated detection remains challenging due to pattern variability and market noise. This paper presents a novel hybrid framework that combines rule-based geometric constraints with Artificial Neural Networks (ANN) for robust Double Top (DT) pattern identification in financial time series. Our methodology employs adaptive thresholding derived from Average True Range (ATR) metrics, swing point prominence analysis, and post-breakout trend confirmation. The ANN component, trained on 16,234 historical patterns across 500 stocks, achieves pattern classification with an accuracy of 88.1% and an F1-score of 82.3%. We introduce a composite validation scoring system integrating geometric fidelity (38%), volume dynamics (27%), breakout quality (22%), and trend confirmation (13%). Experimental results on S&P 500 constituents demonstrate superior performance over baseline methods, with a precision of 83.4%, a recall of 81.2%, and an F1-score of 82.3%, along with a mean absolute error of 3.1% in measured move projections. The framework reduces false positives by 37% compared to traditional rule-based approaches while maintaining computational efficiency suitable for real-time deployment.

Authors

Naimish Vadodariya, Nilesh Gambhava
Darshan University, India

Keywords

Double Top Pattern, Chart Pattern Recognition, Artificial Neural Networks, Algorithmic Trading, ATR-Based Adaptation