Evaluating chaotic functions with flower pollination algorithm for modelling an optimized low complexity neural network based NAV predictor model

Obtaining accurate and impartial forecasts from historical financial data is an emerging and dominant research area in the vast sphere of finance. An investment instrument in the market for structured investments is referred to as a mutual fund, and the tool to measure its price is the Net Asset Value (NAV). Investment managers and investors rely on accurate estimates of NAV of different investment funds to make informed decisions. However, predicting such complex financial series is challenging due to uncertainty and influence of economic and political factors. In this study, a NAV predictor model is designed using a low-complexity neural network, called Legendre Polynomial Neural Network (LPNN). Additionally, a Chaotic Flower Pollination Algorithm (CHFPA) was proposed in the learning phase of the network for tuning its unknown parameters. A series of ten chaotic maps has been applied to the classical FPA through three stagesto achieve improvement in the diversity of the initial population as well as to avoid local optima by carrying out unexpected swaps between global and local search processes. With an aim of accelerating global convergence speed, the selected chaotic variants have been thoroughly investigated on ten well known standard benchmark functions to find the most efficient one. The proposed LPNN-CHFPA NAV predictor model is tested on three separate data sets of reputed Indian financial firms for one day-ahead forecasting of NAV. In comparison with the other prediction models, based on the convergence curve, scatter plots, and various error metrics, the experimental results clearly show that the proposed framework performs well.