Research Article
Detection of Touchscreen-Based Urdu Braille Characters Using Machine Learning Techniques
Table 3
Comparative analysis of Braille to natural language conversion using Naïve Bayes, DT, SVM, KNN, sequential model, and GoogLeNet model.
| Deep learning scheme for character prediction with position-free touchscreen-based Braille input method [33] | Classification techniques used | TPR (%) | TNR (%) | PPV (%) | NPV (%) | TA (%) | FPR (%) |
| Naïve Bayes | NaN | 96.64 | NaN | 99.38 | 96.38 | 3.36 | DT | NaN | 98.82 | NaN | 98.23 | 97.20 | 1.18 | KNN | NaN | 98.60 | NaN | 98.11 | 97.04 | 1.40 | SVM | 68.67 | 99.16 | 76.75 | 98.72 | 83.00 | 0.84 | Sequential model | 90.76 | 99.63 | 91.07 | 99.2 | 92.21 | 0.36 | GoogLeNet model | 95.89 | 99.83 | 96.61 | 99.83 | 95.8 | 0.16 | Detection of Urdu Braille characters based on reconstruction independent component analysis (RICA) features using robust machine learning techniques | DT | 90.98 | 99.78 | 91.21 | 99.78 | 99.57 | 0.22 | KNN | 90.2 | 99.72 | 89.75 | 99.77 | 99.5 | 0.28 | SVM | 93.96 | 99.85 | 94.51 | 99.87 | 99.73 | 0.15 |
|
|