Malaria remains a major global health challenge. Although effective control relies on testing all suspected cases, asymptomatic infections in school-age children are frequently overlooked. Advances in retinal imaging and computer vision have enhanced malaria detection. However, noninvasive, point-of-care malaria detection remains unrealized, partly because of the need for specialized equipment. Here we report radiomic analyses of 4302 photographs of the palpebral conjunctiva captured using unmodified smartphone cameras from asymptomatic 405 participants aged 5 to 15 years to predict malaria risk. Our neural network classification model of radiomic features achieves an area under the receiver operating characteristic curve of 0.76 with 95% confidence intervals from 0.68 to 0.84 in distinguishing between malaria-infected and non-infected cases in endemic regions. Photographing the inner eyelid provides the advantages of easy accessibility and direct exposure to the microvasculature. This mobile health approach has the potential for malaria prescreening and managing febrile illness in resource-limited settings.