Improving BMI Classification Accuracy with Oversampling and 3-D Gait Analysis on Imbalanced Class Data