Table 1 A summary of key research questions and potential research agenda for IR-ML applications relevant to malaria surveys and diagnostics

Incomplete understanding of the IR spectroscopic signals relative to specific biological traits

There is an insufficient understanding of the IR spectroscopic signals (vibrational absorption bands/wavelengths) and their association with biological traits such as parasite infections, age, species, and blood meals

[22, 26, 27, 118]

Inadequate field validation of the IR-ML approaches

There is insufficient field validation of the performance of IR-ML methods for assessing important entomological and parasitological indicators

[22, 23, 28, 120]

Gaps in machine learning frameworks for the IR spectroscopy analysis

There is a need for studies to identify optimal ML objectives such as computational efficiency, prediction accuracy, and model generalizability. This might entail one or a combination of the many existing unsupervised and supervised algorithms

[22, 23, 28, 123]

Unknown detection thresholds

There has not been sufficient demonstration of the limits of detection of IR-ML techniques for detecting malaria infections in human or mosquito samples

[32, 118, 123]

Uncertain granularity of discretized biological outcomes

It is uncertain which method of classifying mosquito age is the best. For example, comparing classification by specific days (1, 2, 3, 4 days) to using longer ranges of days (1, 3, 5, 7 days) or grouping days into ranges (1–5, 5–7, 7–10 days) is unclear

[22, 105]

Resolving overlap and interactions between signals

For biological indicators such as blood meal identification, the possibilities of detecting mixed blood sources remain unknown, and how long after feeding, the blood can still be detected

[22, 105]

Lack of evidence from different epidemiological profiles or settings

There is a need to demonstrate the performance of the IR-ML techniques for detecting malaria parasites in areas with varying epidemiological strata- with low to high transmission or prevalence, and in conditions with varying parasite densities

[27, 28]

Gaps related to hardware and software for IR and ML

There are limited off-the-shelf portable tools that are completely ready for applications in malaria surveys and diagnostics in both laboratory and field settings

[26, 28, 118]

Need to standardize sample-handling procedures

There is currently no standardized protocol for sample handling when using IR-ML methods for malaria surveys and diagnostics

[113, 119, 130]