A novel chaotic flower pollination algorithm for modelling an optimized low-complexity neural network-based NAV predictor model

• Published in: Progress in artificial intelligence 2022-12, Vol.11 (4), p.349-366
• Main Authors: Mohanty, Smita, Dash, Rajashree
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial Intelligence; Chaos theory; Complexity; Computational Intelligence; Computer Imaging; Computer Science; Control; Data Mining and Knowledge Discovery; Evolutionary algorithms; Flowers; Investments; Mathematical models; Mechatronics; Natural Language Processing (NLP); Neural networks; Parameters; Particle swarm optimization; Pattern Recognition and Graphics; Political factors; Polynomials; Regular Paper; Robotics; Vision
• ISSN: 2192-6352; 2192-6360
• DOI: 10.1007/s13748-022-00289-z

Investment instruments for structured investments include mutual funds, and the net asset value (NAV) is used to calculate their value. Due to uncertainty and influences from economic and political factors, it is challenging to predict such complex financial series. The study developed a model to predict NAV by using a low-complexity neural network, the Legendre polynomial neural network (LPNN). Moreover, a new chaotic flower pollination algorithm (NCHFPA) was developed to adjust the unknown parameters of the network through the learning process. NCHFPA is a fusion of chaos-based meta-heuristics with differentiated evolution (DE) algorithm in the local pollination phase of flower pollination algorithm (FPA). In order to determine the best variant of NCHFPA, five different chaotic functions have been investigated in three control parameters. The model was enhanced by integrating the natural evolution features from DE and the pollination process from FPA along with chaos theory. Three real-time mutual fund data sets of reputed Indian financial firms Aditya Birla (AB), SBI and ICICI were used to test this proposed LPNN-NCHFPA model. In order to verify and validate the predictor model further, a comparative analysis is performed with other optimization algorithms such as FPA, Chaotic FPA, particle swarm optimization (PSO) and DE. The proposed framework exhibits an improved performance of 36.65%, 28.22%, 20.10% and 17.18% in RMSE over LPNN-PSO, LPNN-DE, LPNN-FPA and LPNN-CHFPA, respectively, for AB mutual fund. For SBI, an improvement of 46.88%, 32.31%, 18.77% and 6.05% in RMSE and for ICICI, an improvement of 28.87%, 24.58%, 15.63% and 10.05% in RMSE are reported over LPNN-PSO, LPNN-DE, LPNN-FPA and LPNN-CHFPA, respectively, which clearly reveal the competency of the proposed framework over other experimented models.