Kafunda Katalay Pierre
Department of Mathematics, Statistics and Computer Science, Faculty of Science and Technology, University of Kinshasa, Kinshasa, DR Congo

Published : 2 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 2 Documents
Search

Natural Language Processing-Based Financial Time Series Forecasting: Utilizing Sentiment Analysis for Improved Stock Price Prediction Albert Ntumba Nkongolo; Yae Olatoundji Gaba; Kafunda Katalay Pierre; Esther Matendo Mabela; Ben Mbuyi Mpumbu
Journal of Innovation Information Technology and Application (JINITA) Vol 7 No 1 (2025): JINITA, June 2025
Publisher : Politeknik Negeri Cilacap

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35970/jinita.v7i1.2290

Abstract

This study explores the application of natural language processing (NLP) techniques in financial time series forecasting, specifically in predicting stock prices. Historical stock price data and textual data from financial news articles and social media sources were collected, and TextBlob was used to obtain sentiment indices from the textual data. A hybrid model combining NLP techniques with LSTM (Long Short-Term Memory) neural networks was developed, and the methodology involved preprocessing and analyzing textual data using sentiment analysis with TextBlob and integrating the sentiment indices with historical stock price data for forecasting with LSTM. The LSTM model achieved a performance of 89.6 percent precision and outperformed traditional time series forecasting models in terms of accuracy and reliability. The results demonstrate that incorporating sentiment indices obtained through NLP significantly enhances the predictive performance of stock price forecasting models, and the study highlights the potential of NLP techniques, particularly sentiment analysis with TextBlob, in conjunction with LSTM neural networks, to improve the accuracy of financial time series forecasting, specifically in predicting stock prices.   Studi ini mengeksplorasi penerapan teknik pemrosesan bahasa alami (Natural Language Processing/NLP) dalam peramalan deret waktu keuangan, khususnya untuk memprediksi harga saham. Data harga saham historis dan data tekstual dari artikel berita keuangan serta sumber media sosial dikumpulkan, dan TextBlob digunakan untuk memperoleh indeks sentimen dari data tekstual tersebut. Sebuah model hibrida yang menggabungkan teknik NLP dengan jaringan saraf LSTM (Long Short-Term Memory) dikembangkan, dan metodologinya melibatkan praproses dan analisis data tekstual menggunakan analisis sentimen dengan TextBlob, serta integrasi indeks sentimen dengan data harga saham historis untuk peramalan menggunakan LSTM. Model LSTM ini mencapai kinerja dengan tingkat ketepatan (precision) sebesar 89,6 persen dan mengungguli model peramalan deret waktu tradisional dalam hal akurasi dan keandalan. Hasilnya menunjukkan bahwa penggabungan indeks sentimen yang diperoleh melalui NLP secara signifikan meningkatkan kinerja prediktif model peramalan harga saham, dan studi ini menekankan potensi teknik NLP, khususnya analisis sentimen dengan TextBlob, dalam kombinasi dengan jaringan saraf LSTM, untuk meningkatkan akurasi peramalan deret waktu keuangan, khususnya dalam memprediksi harga saham.
Detecting Financial Fraud Using Random Forest Machine Learning Peta Kahiomba Esther; Mabela Matendo Rostin; Kafunda Katalay Pierre; Mbuyi Mukendi Eugene; Mitelezi Mbila Jonathan; Albert Ntumba
Journal of Applied Informatics and Computing Vol. 10 No. 3 (2026): June 2026
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v10i3.12539

Abstract

Financial fraud detection is a critical challenge for banking institutions facing increasingly sophisticated threats in digital transaction environments. This study investigates the application of the Random Forest algorithm for detecting fraudulent credit card transactions using the publicly available benchmark dataset from the Université Libre de Bruxelles (284,807 transactions, 0.172% fraud prevalence). Pre-processing includes QuantileTransformer normalization and SMOTE oversampling applied exclusively to the training set to address class imbalance. The model (n_estimators = 200) is validated using a stratified 70/30 split combined with 10-fold cross-validation to ensure robustness and prevent overfitting. Results yield an accuracy of 97%, ROC-AUC of 97%, precision of 95%, recall of 78%, and F1-score of 86%. Comparative evaluation against Logistic Regression, Support Vector Machine, and Gradient Boosting confirms that Random Forest provides the best balance between detection performance and computational efficiency (training: 45 s; inference: 0.3 ms per transaction). Feature importance analysis identifies transaction amount and PCA components V14 and V17 as the most discriminative variables. Confusion matrix analysis reveals 68 False Negatives and 142 False Positives out of 85,443 test samples. Despite these results, limitations include reduced feature interpretability due to PCA transformation, potential geographic data bias, and real-time production deployment challenges. This work confirms the relevance of Random Forest for financial fraud detection and opens perspectives toward hybrid deep learning and graph-based architectures.