IIUC Journal of Science and Engineering
Permanent URI for this communityhttp://dspace.iiuc.ac.bd/handle/123456789/8468
Browse
2 results
Search Results
Item Arabic character recognition in air-writing based on motion data from 3-axis accelerometer(CRP, International Islamic University Chittagong, 2023-12) Kader, Mohammed AbdulAir-writing is a cutting-edge, non-touch human-machine interaction technique that enables users to input text into digital devices by moving their hands in free space, instead of using conventional input devices like keyboards and touch screens. This approach appears to be one of the most effective ways to enter text into digital systems in the future. English air-writing has received significant scholarly attention, but no studies on Arabic air-writing were found. In this research, a system is developed to recognize Arabic characters in air-writing based on motion data from a 3-axis accelerometer. A data acquisition system is constructed to record hand movements during air-writing. Each Arabic letter is written 25 times in the air using this data acquisition system, and a motion-sensor-based Arabic air-writing dataset is prepared. Using this dataset, several supervised machine learning models have been trained, and their accuracy has been determined. It is observed that the Fine KNN and Quadratic SVM models have demonstrated the highest accuracy (98.5%) in identifying Arabic characters from air-writing among the various available supervised machine learning models.Item Unleashing class imbalance problem in loan dataset through a novel oversampling approach based on FCM(CRP, International Islamic University Chittagong, 2023-12) Akter, SubrinaCommercial companies highly depend on loan approval models trained by machine learning and statistical methods to predict loan status. However, imbalanced datasets present a key challenge in this sector. Addressing this issue, this paper proposes a new oversampling method based on Fuzzy C means clustering. This clustering algorithm assigns the instances to several groups by assigning a flexible degree of membership. K-Nearest Neighbor and the Decision Tree served as the basic classifiers in an extensive test on the Kaggle loan dataset. Three distinct imbalanced ratios—2.2, 4.2, and 8.45—were used in the experiment. The effectiveness of the recommended strategy was compared with SMOTE and WBOT using 5-fold CV. The outcomes showed that the proposed method outperformed both SMOTE and WBOT, obtaining higher average F-measure and G-mean values across the machine learning algorithms. These findings show how the suggested method may correct class imbalance while also enhancing prediction accuracy in the context of loan acceptance.