<![CDATA[Due to intense volatility, complex nonlinear dynamics, and scant historical data, predicting financial prices in emerging markets is extremely difficult. This paper presents a hybrid ARIMA+LSTM model for stock price forecasting in the Indonesian market and tests it. The model effectively combines traditional econometric techniques with advanced deep learning methodologies. Walk-forward validation on over five years of data from various Indonesian stocks (BBRI, ALFMART, UNVR, BSIM) is applied. The hybrid model achieves a Root Mean Square Error of 112.54, Mean Absolute Percentage Error of 2.21%, and Directional Accuracy of 68.9% for one-day ahead predictions. This performance exceeds that of pure ARIMA by 22.5% and is statistically significant (p < 0.001, Cohen's d = 1.18). The model consistently shows good results over many prediction horizons (1, 5, and 10 days) and several Indonesian stocks from different sectors with a standard deviation of only 8.3 during the test period. A cloud-based deployment architecture is planned to reach about 1,500 predictions per second at a latency of 45ms which will be suitable for real-time institutional trading systems. Sensitivity analysis reveals optimal hyperparameters (60-day window; between 50 to 25 LSTM units with a dropout rate of 0.2) as well as confirming strong performance across parameter variations. SHAP analysis plus attention visualization results show that the model keeps interpretability even though deep learning is complicated; recent prices (lag-1 and lag-2) hold about 70% of the prediction variance. This work validates hybrid ARIMA+LSTM modeling in an emerging market like Indonesia through rigorous walk-forward validation methodology and practical insights into generating actionable trading signals with a win rate of 68.9% which supports portfolio management integrated within risk frameworks as well as limitations that include dependency on historical data, exclusion of transaction costs, and single asset focus yet significantly contributes methodological rigor and empirical validation to machine learning literature in financial forecasting specifically regarding emerging markets]]>
