<![CDATA[This study develops a Hybrid Decision Support System (DSS) for structured stock screening of all 45 LQ45 index constituents on the Indonesia Stock Exchange, addressing the absence of a unified analytical tool that integrates fundamental valuation, technical momentum, and news sentiment within a single score-based framework accessible to retail investors. The proposed system integrates three independent analytical modules: an Auto Valuation Engine employing six intrinsic valuation methods with dynamic sector weighting for fundamental analysis; a Soft Voting Ensemble of Random Forest, XGBoost, and Logistic Regression trained on 32 technical and 6 fundamental features for technical classification; and a hybrid IndoBERT architecture augmented with a domain-specific financial lexicon (60/40 weighting) for real-time sentiment analysis of Indonesian-language financial news. A weighted Final Score (50% Fundamental, 30% Technical, 20% Sentiment) is computed as a unified quantitative representation of each constituent's condition on a 0–100 scale, with model evaluation conducted using a chronological 80:20 hold-out split and 5-fold TimeSeriesSplit cross-validation across all 45 constituents, supplemented by a walk-forward portfolio backtest (2 June–7 September 2026; 4 rebalancing cycles; transaction costs included) in which the Quantamental Top-5 portfolio achieved a total return of +7.15% with a Sharpe Ratio of 1.01 and Maximum Drawdown of −9.00%, outperforming the IHSG benchmark (+6.85%, Sharpe 0.78, MDD −13.77%) and the Technical ML-only strategy (−10.22%), demonstrating the value of multi-pillar integration for risk-adjusted screening. The Soft Voting Ensemble outperformed individual classifiers on three of five metrics Accuracy (53.84%), Macro Precision (52.48%), and F1 Weighted (50.01%) while Logistic Regression achieved the highest F1 Macro (48.39%), as detailed in Table 5. IndoBERT independently produced a uniform neutral score (50.00) for 82.2% of constituents, empirically confirming domain mismatch and validating the necessity of a hybrid lexicon correction mechanism. Final Score distribution ranged from 16.86 to 64.17 with a mean of 42.57 across all 45 LQ45 constituents. To the best of the authors' knowledge, and based on a systematic review of prior LQ45 and Indonesian stock DSS literature, no identified study integrates all three analytical pillars within a unified, continuously scored DSS applied to the complete LQ45 universe; this study addresses that gap, as substantiated in Table 1.]]>
Copyrights © 2026
