Table 9.

The future challenges reported in this domain.

Reference	Category	Discussed Future Challenge
[60]	Limited public data set	Since public data sets are not accessible, non-public databases and photographs gathered via the Internet are used for research. This complicates the replication of the findings since the dataset is not available.
[26]	Light-colored skin images	Since 2016, ISIC has arranged an annual melanoma diagnostic competition, but the presence of only light-skinned data is one of the drawbacks of ISIC. Dark-haired photographs are needed to be included in the datasets.
[21]	Lesion size impact on training	The lesion scale has also been found to be significant in most studies if the lesion scale is smaller than 6 mm, so melanoma cannot be identified, and the sensitivity of the diagnosis falls significantly.
[57]	Deep learning accuracy improvement	Deep learning approaches have been found to work correctly for 70% of training pictures and 30% of testing pictures. In comparison, findings require that the training ratio is necessary for good outcomes. The deep learning methods work well where the optimum balance is set. It is challenging to devise hybrid strategies that can work better with fewer training ratios.
[26]	Non-availability of fusion methods	Most of the techniques are focused on basic deep learning methods. However, fusion techniques are reported with better accuracy. Despite this, the fusion techniques are less reported in the literature for specific data sets.