Time Series Mining Approaches for Malaria Vector Prediction on Mid-Infrared Spectroscopy Data

Abstract

Malaria is an infectious disease caused by the Plasmodium parasite transmitted to humans by the bite of infected female Anopheles mosquitoes. The disease remains a major cause of child mortality globally and caused more than 600,000 deaths in 85 countries only in 2022, predominantly affecting the African region. Recent discussions point out that climate change is expanding the geographical distribution of mosquitoes, accelerating the malaria burst in regions previously free from outbreaks. Traditional vector control relies on chemical methods (e.g., insecticides), but effective control implementation requires accurate and cheap mosquito population monitoring and longevity estimates. This study investigates using mid-infrared spectroscopy (MIRS) data as input for efficient time series classification methods to predict the species and age of malarial mosquitoes. Unlike previous studies using traditional machine learning algorithms, our comprehensive evaluation includes 14 algorithms from four time series mining approaches, such as feature-based, interval-based, convolutional-based, and deep learning methods. These methods consider the particularities of time series presented in MIRS data, such as temporal dependencies and correlations between features. Results indicate that the deep learning algorithm InceptionTime achieves 97% species identification accuracy and 83% age prediction accuracy, outperforming the traditional methods evaluated in the literature. This research contributes to the field by highlighting the effectiveness of time series mining approaches for malaria vector control using spectroscopy. As malaria continues to pose a significant threat, these advancements contribute to developing innovative and efficient tools for malaria control strategies.

Keywords

• mosquitoes
• malaria
• time-series classification
• mid-infrared spectroscopy
• machine learning
• deep learning