Improving Stock Price Prediction through a Multilayer Perceptron Driven by a Grasshopper Optimization Algorithm: An Analysis of the Hang Seng Index

Abstract

Given that time-series data is nonlinear, noisy, and dynamic, it may be difficult to predict stock prices in turbulent financial markets. To tune essential MLP hyperparameters such as the number of hidden units, learning rate, batch size, and epochs, this study presents a hybrid prediction model that combines a Multilayer Perceptron (MLP) with the Grasshopper Optimization Algorithm (GOA). On an 80/20 train–test split, the model is trained and tested on daily OHLC price and volume data from January 2015 to June 2023 for the Hang Seng Index (HSI). Benchmark models such as Transformer, Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), Outlier-Robust Extreme Learning Machine (OR-ELM), Histogram-Based Gradient Boosting Regression (HGBR), and other hybrid optimizers (BBO-MLP, GA-MLP) are used for comparing performance. With an