Enhancing the chimp optimization algorithm to evolve deep LSTMs for accounting profit prediction using adaptive pair reinforced technique

Abstract Accurately predicting accounting profit (PAP) plays a vital role in financial analysis and decision-making for businesses. The analysis of a business’s financial achievements offers significant insights and aids in the formulation of strategic plans. This research paper focuses on improving the chimp optimization algorithm (CHOA) to evolve deep long short-term memory (LSTM) models specifically for financial accounting profit prediction. The proposed hybrid approach combines CHOA’s global search capabilities with deep LSTMs’ sequential modeling abilities, considering both the global and temporal aspects of financial data to enhance prediction accuracy. To overcome CHOA’s tendency to get stuck in local minima, a novel updating technique called adaptive pair reinforced (APR) is introduced, resulting in APRCHOA. In addition to well-known conventional prediction models, this study develops five deep LSTM-based models, namely conventional deep LSTM, CHOA (deep LSTM-CHOA), adaptive reinforcement-based genetic algorithm (deep LSTM-ARGA), marine predator algorithm (deep LSTM-MPA), and adaptive reinforced whale optimization algorithm (deep LSTM-ARWOA). To comprehensively evaluate their effectiveness, the developed deep LSTM-APRCHOA models are assessed using statistical error metrics, namely root mean square error (RMSE), bias, and Nash–Sutcliffe efficiency (NSEF). In the validation set, at a lead time of 1 h, the NSEF values for LSTM, LSTM-MPA, LSTM-CHOA, LSTM-ARGA, LSTM-ARWOA, and deep LSTM-APRCHOA were 0.9100, 0.9312, 0.9350, 0.9650, 0.9722, and 0.9801, respectively. The results indicate that among these models, deep LSTM-APRCHOA demonstrates the highest accuracy for financial profit prediction.