Nugrahasyach, Muhammad Rafli (2026) Implementasi Explainable AI Pada Peramalan Harga Saham Sektor Teknologi Dan Keuangan Di Indeks LQ45 Menggunakan LSTM Dengan Mekanisme Attention. Diploma thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 2043221085-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
2043221085-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy |
Abstract
Pertumbuhan signifikan jumlah investor ritel di pasar modal Indonesia yang didominasi oleh generasi muda belum diimbangi dengan tingkat literasi pasar modal yang memadai. Kondisi ketimpangan ini menyebabkan berbagai keputusan investasi sering kali didasarkan pada bias emosional seperti Fear of Missing Out (FOMO) yang berkontribusi secara langsung pada tingginya angka kerugian investor ritel. Penelitian ini mengusulkan sebuah solusi berbasis artificial intelligence untuk menyediakan alat bantu keputusan objektif melalui peramalan harga saham di sektor teknologi dan keuangan. Metodologi yang digunakan adalah pendekatan fusi data multimodal berbasis LSTM dengan mekanisme attention yang mengintegrasikan data kuantitatif harga historis dengan data teks berita serta sentimen pasar dari platform Stockbit yang diperoleh dengan web scraping. Data dianalisis menggunakan IndoSBERT untuk relevansi semantik dan model IndoRoBERTa untuk klasifikasi sentimen. Mekanisme attention berfungsi menimbang pengaruh antar-modalitas secara dinamis, sementara metode Explainable AI (XAI) dengan SHAP digunakan untuk menginterpretasikan faktor pendorong utama di balik hasil prediksi. Hasil penelitian menunjukkan model fusion unggul secara signifikan pada saham dengan likuiditas tinggi seperti BBCA dibandingkan model teknikal murni. Sebaliknya, pada saham dengan volatilitas ekstrem seperti GOTO, model baseline teknikal terbukti lebih efektif dalam meredam noise informasi sentimen. Secara keseluruhan, pendekatan multimodal berhasil meningkatkan akurasi serta transparansi model peramalan sebagai instrumen pendukung keputusan investasi taktis yang lebih objektif di tengah tingginya dinamika pasar.
=======================================================================================================================================
The significant growth in the number of retail investors in the Indonesian capital market, which is dominated by the younger generation, has not been balanced by an adequate level of capital market literacy. This disparity causes various investment decisions to be frequently based on emotional biases such as Fear of Missing Out (FOMO), which directly contributes to the high rate of losses among retail investors. This research proposes an artificial intelligence-based solution to provide objective decision support tools through stock price forecasting in the technology and financial sectors. The methodology used is a multimodal data fusion approach based on LSTM with an attention mechanism that integrates quantitative historical price data with news text data and market sentiment from the Stockbit platform obtained through web scraping. Data were analyzed using IndoSBERT for semantic relevance and the IndoRoBERTa model for sentiment classification. The attention mechanism functions to dynamically weigh cross-modal influences, while the Explainable AI (XAI) method with SHAP is used to interpret the main driving factors behind the prediction results. The results showed that the fusion model outperformed the pure technical model for high-liquidity stocks such as BBCA. Conversely, for stocks with extreme volatility such as GOTO, the technical baseline model proved to be more effective in mitigating the noise of sentiment information. Overall, the multimodal approach succeeded in increasing the accuracy and transparency of the forecasting model as an objective tactical investment decision support instrument amidst high market dynamics.
Actions (login required)
 |
View Item |