Strokes constitute major units that form the contents of any handwritten text. When every user feeds or writes any handwriting sample, we archive strokes depending upon the context. In this study, the authors have experimented with recognizing handwritten mathematical expressions on datasets with varying strokes. The case study on mathematical expressions has been conducted where distinct datasets are analyzed, identified, and compared. A deep neural network-based recognizer has been deployed to test the formulated hypothesis around stroke characteristics. One experimented dataset had handwritten samples that had uniformly thin width of the stroke, whereas the other had strokes of varying width. After the experimentation, it has been observed that the dataset with thin stroke width (up to 1px) results in a significantly less effective recognition rate. The one achieved on experimenting with a dataset with varying strokes outcomes a jaw dropping recognition rate compared to the previous case. A comparative analysis has been performed to infer and affirm the hypothesis about stroke characteristics and their impact on the recognition rate.