In Bangladesh, chronic arsenic exposure through drinking water has affected millions of individuals and remains a public health issue. Due to the carcinogenic effects after long-term or high-dose arsenic exposure to multiple organs and systems in the human body, affected individuals are subject to an increased risk for non-melanoma skin cancer and other cancers. While the effort to eradicate this contamination is inconclusive due to unsustainability, complexity, potentially greater secondary pollution, and other drawbacks, oxidative stress and DNA damage have been suggested and hypothesized to underlie arsenic carcinogenesis. To develop well-rounded prevention strategies, we divided the goal into two parts: first, through lowering the impact from the exposure to lower the disease risk; and second, through early detection to minimize morbidity and mortality. The primary prevention strategy for nonmelanoma skin cancer (NMSC) was carried out through intervening in the disease. In cancer prevention, the blood levels of antioxidants, namely vitamin E and selenium, have demonstrated protective effects by combating free radicals in our body. This mechanistic evidence set up a foundation for us to explore therapeutic options for the affected populations and their subgroups. In the first study of this dissertation (Chapter 1 and Chapter 2), I evaluated the overall and differential treatment effects of selenium and vitamin E, alone or in combination on the risk of NMSC and mortality outcomes. There were no statistically significant overall and treatment effects observed on both endpoints of the trial. Additionally, we did not observe differential treatment effects by ten baseline population characteristics, including gender, BMI, smoking status, sun exposure, occupation, skin lesion severity, urinary arsenic level, blood selenium, plasma α-tocopherol, and plasma γ-tocopherol. The second approach to reduce morbidity and mortality due to NMSC was through early and accurate recognition of skin cancer. In the second part of this dissertation (Chapter 3), I applied an attention-based deep neural network to analyze histopathological images collected from the participants who were suspected of NMSC. Our model showed promising diagnostic accuracy for non-cancer and basal cell carcinoma subtypes. Moreover, we also generated the heatmaps from the model and visualized the key areas within each image that drive the cancer diagnosis. These machine-generated heatmaps were proven accurate in pinpointing the lesions by the expert dermatopathologist. In summary, we conducted a thorough evaluation to alleviate the impact of NMSC for the susceptible population in Bangladesh through both treating the disease as well as making efforts for early and more accessible cancer detection procedures. These prevention strategies would serve as a model for the similar frameworks of other cancers and complex diseases in a similar setting.