Introduction

Dengue fever has remained as the preeminent disease in Indonesia  for nearly six decades, establishing the country’s stature as a global epicentre of the disease due to its conducive warm and humid  climate that fosters ideal conditions for mosquito proliferation and  viral transmission. Despite availability of treatments, the field of microbiology predominantly focuses on designing potent drugs, causing research to overlook a critical aspect crucial to achieve optimal healthcare outcomes: time  of diagnosis. While the refinement of effective drugs is undoubtedly valuable, there exist a gap in the consideration of timely diagnosis that impedes such drugs from achieving optimal efficacy. In addition, treatment for Dengue patients is typically administered  post-manifestation of visible symptoms, thereby elevating the risk of  infections progressing to Dengue hemorrhagic fever (DHF) or  Dengue shock syndrome (DSS).  Current methods of manual identification  and laboratory tests of aedes aegypti are  likewise inefficient and time-consuming;  requiring specialized equipment and  expertise.  This project aimed to redirect efforts through utilizing unique  characteristics of disease-carrying micro-organisms (Aedes Aegypti)  for an efficient approach towards early detection of diseases (Dengue  Fever) via binary classifiers.

Experiment Methods

A. Software Process:

1.  Audio Pre-processing: all audio files were standardized with the same frequency and sample rate. The Audacity software was used to apply the low pass filter and downsampling. During audio segementation, the files were divided into smaller overlapping segments. An audio segment of 1 window size W seconds was copied. It was then processed and the placeholder of the audio file by slide size S =2 was advanced. 

2. Audio Segementation: the audio segements were transformed into recognizable features for the neural network in the form of spectrograms. Each spectrograms was labelled for supervised training.

3. Feature Extraction: the audio segment was then decomposed into a spectrogram through the generate spectrogram function, which applies the Short-Term Fourier Transform (STFT) to extract features. 

B. Spectrograms: each species exhibited distinct harmonic distributions; notably Aedes aegypti possessed a higher fundamental frequency compared to other mosquitoses.

C. Epoch: The dataset was split into data data teestingand training. The training familirized the model with the dataset to thus allow it to adjust its weight better. Epoch 4 had the highest accuracy of 92%.

D. User Interface: Swift on Xcode was used. The initial page provides an option for users to upload an existing wav file or record their surroundings. Afterwards, they are directed to the second page that provides a percentage for the likelihood of dengue along with the spectrogram. They can then choose to learn preventive measures they could apply in their area to reduce this percentage. 

Conclusion, Application and Future Outlook

The application can be utilized by individuals across Indonesia and allow them to implement preventive  measures, particularly in regions with high dengue endemicity. This would allow them to mitigate the impact  of dengue outbreaks, and hopefully it can contribute to  improved public health outcomes against mosquito borne diseases. For future work, as this application is currently only available in Apple devices, it could be  expanded to Android. Furthermore, the dataset could be expanded and advanced to the development of  neural networks to update with advances in the area. Neural networks could also be made resistant to noise  generated deliberately by malicious users attempting to induce false positives.