Machine Learning Approaches for Predicting Mobile Financial Services Growth: A Comprehensive Multi-Target Analysis of Bangladesh's Digital Financial Ecosystem

Abstract Mobile Financial Services (MFS) have catalyzed unprecedented financial inclusion in developing economies, with Bangladesh exemplifying transformative digital financial innovation. This study presents the first comprehensive multi-target machine learning framework for predicting key MFS growth indicators, addressing critical gaps in financial technology forecasting literature. Using 76 months of comprehensive Bangladesh Bank data from December 2018 to March 2025, spanning 24 variables, we systematically evaluate nine machine learning algorithms across three paradigms: linear models, tree-based ensembles, and deep learning architectures. Our rigorous methodology encompasses advanced feature engineering, temporal pattern recognition, and target-specific optimization. Results reveal unprecedented predictive accuracy with target-specific algorithmic superiority: Ridge Regression achieves exceptional performance for transaction count prediction with R² = 0.9978, RMSE = 6.76M transactions, LSTM networks demonstrate superior capability for transaction amount forecasting with R² = 0.9926, RMSE = 32,486M BDT, and Temporal Convolutional Networks excel in float amount prediction with R² = 0.9726, RMSE = 5,674M BDT. Feature importance analysis identifies temporal dependencies and transaction type diversity as primary growth drivers. These findings establish new benchmarks for financial service prediction while providing actionable intelligence for policymakers, financial institutions, and fintech innovators.