Item-Based Collaborative Filtering Recommendation Engine for Adaptive Financial Literacy Enhancement in Mobile Learning Environment

The rapid growth of digital investment platforms in Indonesia has not been matched by adequate financial literacy, exposing novice investors to significant risks. This study addresses this problem by developing a mobile-based adaptive learning system that implements an optimized Item-Based Collaborative Filtering (IBCF). The novelty of this system lies in the integration of a hybrid cold-start mitigation strategy combining popularity metrics with TF-IDF content-based filtering to maintain recommendation accuracy for new users. The application was developed using the Waterfall model with Flutter framework for cross-platform interface and Firebase for serverless backend architecture. The IBCF algorithm employs cosine similarity for computing item-item relationships and weighted sum method for generating prediction scores. System evaluation utilized a mixed-method approach: algorithmic performance was measured using Root Mean Square Error (RMSE) and Precision/Recall metrics, usability was assessed via System Usability Scale (SUS), and educational effectiveness was analyzed through Paired Sample T-Test on Pre-Test and Post-Test scores. Results demonstrate that the recommendation engine successfully adapted to user patterns, evidenced by RMSE reduction from 0.98 to 0.88 after field testing, achieving final Precision of 7.65% and Recall of 19.51%. The system attained a SUS score of 75.95 (categorized as "Good"), and significantly improved user investment knowledge with mean scores increasing from 46.03 to 58.23 (p < 0.001). This research validates that the proposed IBCF-based adaptive system outperforms baseline models, including Content-Based Filtering (RMSE: 1.0114), User-Based Collaborative Filtering (RMSE: 1.0333), and non-personalized approaches (RMSE: 1.1235), achieving superior predictive accuracy with an RMSE of 0.8832 in high-risk investment contexts.