Ahmed, Salman (2025) Forecasting the Prices of diverse cryptocurrencies using Machine Learning, Deep Learning, and Explainable AI with Hybrid Sentiment Analysis from News Articles and Financial Market Data. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
The cryptocurrency market is one of the most volatile financial spaces, mainly attributed to its exposure to speculative sentiments, lack of regulatory oversight, and unpredictable pricing factors. This study addresses the challenge of forecasting price movements by analysing ten major cryptocurrencies ETH, BTC, ADA, SOL, XRP, AVAX, LINK, DOT, LTC, and TRX using a combination of financial indicators and hybrid sentiment features derived from multiple online sources. The market data was collected via the Kraken API, while news and discussion-based sentiment signals were collected from Reddit, GDELT, and Mediastack using VADER, TextBlob, and FinBERT models. The derived sentiment scores were normalized, aggregated, and combined with technical indicators to formulate an integrated feature set. This feature set was then used to train six different machine learning models: Random Forest, Extreme Gradient Boosting, CatBoost, LightGBM, HistGradientBoosting, and Multi-Layer Perceptron Regressor. The model performances were evaluated using RMSE, MAE, and R² for each asset. Among the models analysed, tree- based ensemble methods namely LightGBM and CatBoost showed consistently better predictive power, with various models registering R² values greater than 0.98. To make the models more interpretable and support decision-making processes, Explainable Artificial Intelligence (XAI) techniques were applied. SHAP (Shapley Additive Explanations), Partial Dependence Plots (PDP), and Individual Conditional Expectation (ICE) were used to visualize and quantify the contribution of hybrid sentiment, volatility, and volume-based features for different assets. The results confirm that hybrid sentiment significantly enhances short-term forecasting accuracy, particularly for altcoins with considerable social influence. The proposed framework demonstrates that the combination of multisource sentiment analysis and explainable machine learning allows for more transparent, interpretable, and accurate forecasting in complex digital asset markets.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Simiscuka, Anderson UNSPECIFIED |
| Subjects: | Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing H Social Sciences > HG Finance > Money > Digital currency Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Artificial Intelligence |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 11 Aug 2026 14:40 |
| Last Modified: | 11 Aug 2026 14:40 |
| URI: | https://norma.ncirl.ie/id/eprint/9493 |
Actions (login required)
![]() |
View Item |
Tools
Tools