Artiﬁcial Intelligence-Based Detection and Prediction of Giant African

Snail (Lissachatina Fulica) Infestation in the Gal

apagos Islands

Jonathan Loor

1 a

, Ariana Jim

enez

1 b

, Juan David Moromenacho Aguirre

2 c

, Grace Rodr

ıguez

3 d

an Reyes

2 e

, Paulina Vizcaino-Imaca

2 f

and Manuel Eugenio Morocho-Cayamcela

1,2 g

Yachay Tech University, School of Mathematical and Computational Sciences, DeepARC Research Group,

Hda. San Jos

e s/n y Proyecto Yachay, Urcuqu

ı, 100119, Ecuador

Universidad Internacional del Ecuador, Faculty of Technical Sciences, School of Computer Science,

Quito, 170411, Ecuador

Pontiﬁcia Universidad Cat

olica del Ecuador, Faculty of Exact and Natural Sciences Sciences, Biology,

Quito, 170525, Ecuador

Keywords:

Pest Management, Invasive Species Detection, Biodiversity Conservation, Gal

apagos Islands, Artiﬁcial

Intelligence, Transformer time-series, Mobile Application Development.

Abstract:

The Gal

apagos Archipelago are confronting a signiﬁcant threat from invasive species, notably L. fulica, which

disrupts the delicate balance of their natural ecosystem. An innovative solution is proposed, employing mobile

application technology and artiﬁcial intelligence (AI) to streamline the collection, analysis, and prediction of

L. fulica movements. The mobile application facilitates efﬁcient recording of L. fulica sightings by ﬁeld teams,

including Global Positioning System (GPS) coordinates, type, condition, and quantity. Data collected is trans-

mitted to a cloud-based server for storage and analysis, where machine learning algorithms process time-series

data to generate predictive models of L. fulica movement patterns. Results underscore the effectiveness of AI

in enhancing the efﬁciency and accuracy of Giant African Snail (GAS) detection and movement estimation,

facilitating informed decision-making by administrators and managers. By safeguarding the native ﬂora and

fauna of the archipelago, this solution represents a signiﬁcant stride towards mitigating the impact of invasive

species and preserving the unique biodiversity of the Galapagos Islands.

1 INTRODUCTION

The Gal

apagos Archipelago, renowned for its unpar-

alleled biodiversity and unique ecosystems, faces an

ongoing challenge in the form of invasive species

that threaten its delicate ecological balance (Khatun,

2018), (Collins et al., 2019). Among these invaders,

Lissachatina fulica, the giant African snail stands out

as a particularly serious threat (Miquel and Herrera,

2014). This mollusk ranks high on the list of the

100 most harmful alien species in the world due to its

rapid proliferation and devastating impact on native

https://orcid.org/0009-0003-0802-0858

https://orcid.org/0009-0002-4838-1538

https://orcid.org/0009-0007-6014-8911

https://orcid.org/0009-0006-8380-1306

https://orcid.org/0009-0002-2731-5531

https://orcid.org/0000-0001-9575-3539

https://orcid.org/0000-0002-4705-7923

ﬂora and fauna (Simberloff and Rejmanek, 2020)-

Traditional methods of monitoring and controlling L.

fulica infestations rely heavily on manual data collec-

tion processes, which are often labor-intensive, time-

consuming, and prone to errors (Elias, 2022). In re-

sponse to these limitations, there is an urgent need

for innovative approaches that leverage technology to

enhance the efﬁciency and effectiveness of pest detec-

tion and management efforts.

In this paper, we present a novel solution aimed

at automating the detection and prediction of L. fulica

infestations in the Gal

apagos Archipelago through the

integration of mobile application technology and arti-

ﬁcial intelligence (AI). Our proposed system seeks to

revolutionize the way L. fulica sightings are reported,

recorded, and analyzed, thereby facilitating proactive

and data-driven decision-making in pest control oper-

ations.

Our solution leverages mobile devices and AI

algorithms to empower ﬁeld squads conducting L.

Loor, J., Jiménez, A., Aguirre, J., Rodríguez, G., Reyes, I., Vizcaino-Imacaña, P. and Morocho-Cayamcela, M.

Artiﬁcial Intelligence-Based Detection and Prediction of Giant African Snail (Lissachatina Fulica) Infestation in the Galápagos Islands.

DOI: 10.5220/0012763200003753

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 19th International Conference on Software Technologies (ICSOFT 2024), pages 403-410

ISBN: 978-989-758-706-1; ISSN: 2184-2833

403

fulica surveillance by enabling real-time data collec-

tion and transmission directly from the ﬁeld. The mo-

bile app streamlines data collection and automatically

extracts temperature and humidity data using Google

APIs, aiding in pest classiﬁcation and seasonal anal-

ysis. This reduces the workload on ﬁeld personnel

while enhancing data accuracy and timeliness. The

centralized server hosted in the cloud receives and

analyzes collected data using sophisticated machine

learning algorithms to generate predictive models of

L. fulica movement patterns. These models, based

on historical data and environmental variables, allow

stakeholders to anticipate and proactively respond to

potential outbreaks, minimizing the impact on native

ecosystems.

This represents a signiﬁcant advancement in the

ﬁeld of pest management, offering a scalable and

cost-effective approach to monitoring and controlling

invasive species in sensitive ecological environments.

Through the synergistic integration of mobile technol-

ogy, machine learning, and deep learning, we aim to

empower conservationists and policymakers with the

tools and insights needed to safeguard the unique bio-

diversity of the Gal

apagos Islands for generations to

come.

2 RELATED WORKS

2.1 Plagues and the Gal

apagos

Ecosystem

Pests pose a multifaceted threat by interfering with the

normal development of ecosystems, affecting agricul-

ture, public health, ecology, economy and food secu-

rity (Warner, 2019). This challenge is especially evi-

dent in the Galapagos Islands, where invasive species

pose a crucial threat to the conservation of the biodi-

versity of this natural laboratory. The introduction of

exotic species can lead to the extinction of endemic

species and affect key economic sectors, highlight-

ing the urgency of preventive measures and effective

management. The uniqueness of the Galapagos as an

evolutionary laboratory underscores the need for in-

ternational cooperation and continued research to ad-

dress the emerging dynamics of biological invasions.

One example of this problem is the snail L. fulica,

whose impact on native biodiversity has been docu-

mented. Studies have documented the impact of the

snail L. fulica on native biodiversity, causing the dis-

appearance of endemic snails in the Hawaiian Islands

and economic problems in Nepal due to crop invasion

(Budha and Naggs, 2008)(Cowie, 1998). The high re-

productive rate of L. fulica, with the ability to lay up

to 1000 eggs during its life cycle and reach the repro-

ductive stage in six months, contributes to its invasive

success (Miquel and Herrera, 2014).

Control strategies for L. fulica include methods

such as manual removal, trapping, physical barriers

and the use of chemicals or predators such as Eug-

landina rosea (Gerlach et al., 2021). However, these

methods may have limitations and compromise local

biodiversity (Gerlach et al., 2021). The choice of

strategies must be carefully considered for effective

and sustainable pest management.

2.2 Traditional Human-Based Methods

of Pest Surveillance

Historically, pest surveillance and monitoring in eco-

logical environments have relied on manual methods

such as visual inspections, trapping, and ﬁeld surveys.

While these approaches have provided valuable in-

sights into pest populations and distributions, they are

often labor-intensive, time-consuming, and prone to

human error. Moreover, they may lack the scalability

and real-time data acquisition capabilities needed to

effectively address dynamic pest threats.

Traditional pest surveillance methods, such as vi-

sual inspections, trapping and ﬁeld surveys, have

evolved over time (McCallum et al., 2021). Although

they provide valuable information on pest popula-

tion dynamics and distribution, they face signiﬁcant

challenges, such as labor intensity, time required and

susceptibility to human error(Awuor et al., 2019).

In the Galapagos Islands, visual monitoring, search

and egg collection with subsequent incineration, and

controlled ﬁre are used to address these challenges

(Correoso, 2006). These limitations have led to the

exploration of innovations in pest monitoring, with

the aim of improving efﬁciency and providing a sound

basis for preventive management and pest control in

sensitive environments such as the Galapagos Islands.

2.3 Mobile Applications for Pest

Management

The integration of mobile technology into pest man-

agement practices has emerged as a promising ap-

proach to overcome the limitations of traditional

surveillance methods. Several studies have explored

the development and deployment of mobile applica-

tions for the collection, storage, and analysis of pest-

related data. For example, apps such as PestMapper

and EDDMapS allow users to report sightings of in-

vasive species and contribute to crowd-sourced moni-

toring efforts. Similarly, in Kenya, an innovative pest

ICSOFT 2024 - 19th International Conference on Software Technologies

404

management solution is proposed, focusing on farm-

ers. This solution utilizes mobile devices to process

images and leverages crowdsourcing to assist farm-

ers in effectively identifying and controlling pest inva-

sions in their crops (Vanegas et al., 2018). While these

applications have demonstrated the potential to en-

hance data accessibility and community engagement,

they often lack advanced features for automated im-

age analysis and predictive modeling.

2.4 Transformer time-series in Pest

Detection

Recent advances in transformer-based architectures

have paved the way for automated pest detection us-

ing prognostic analysis techniques. Researchers have

developed algorithms capable of knowing the behav-

ior of an agent and predicting the next destination.

The model is trained for the exact coordinates of lon-

gitude and latitude to predict, allowing for rapid and

accurate identiﬁcation of invasive species. For exam-

ple, by knowing the timing of pests, resources could

be provided to prevent their increase and subsequent

elimination from the environment. However, the ap-

plication of these techniques to real-world pest man-

agement scenarios, particularly in remote or resource-

limited environments, remains relatively unexplored

due to missing data (Abideen et al., 2021).

In this scenario, the team utilizes the PyTorch frame-

work for their project, and they are tasked with cal-

culating position encoding. To deﬁne this mathemat-

ically, they ﬁrst establish the concept of position en-

coding within the context of neural network architec-

tures. Position encoding is a technique commonly

employed in sequence modeling tasks, such as nat-

ural language processing and time series analysis, to

incorporate positional information into the input em-

bedding. Mathematically, the position encoding func-

tion can be deﬁned as follows:

For a given position p in the sequence and dimension

d of the embedding, the positional encoding is com-

puted as:

PE(2, i) = sin



10000



(1)

PE(2, i + 1) = cos



10000



(2)

Where i is the dimension index. These sinusoidal

functions generate values between -1 and 1 and en-

sure a unique and repeatable pattern for each position.

Also, the model is constructed to accept the following

parameters.

• The dimension of the input data, in this case, we

use only one input for each of the coordinates for

the latitude and longitude.

• The number of features in the transformer model’s

internal representations (also the size of embed-

dings). This controls how much a model can re-

member and process.

• The number of attention heads in the multi-head

self-attention mechanism.

• The number of transformer encoder layers.

dropout: The dropout probability.

In the training step batched training ensures that

the model updates its weights based on the aver-

age gradient over several data points, rather than be-

ing excessively inﬂuenced by any single instance.

Also, we can deﬁne early stopping to avoid overﬁt-

ting.Thus, while transformer architectures introduce

novel mechanisms and complexities, the foundational

principles of training deep learning models in Py-

Torch remain consistent. (Heaton, 2024)

2.5 Integrated Approaches to Pest

Management

A growing body of research advocates for integrated

pest management (IPM) strategies that combine mul-

tiple techniques and technologies to achieve more ef-

fective and sustainable pest control outcomes. These

approaches often incorporate elements of biologi-

cal control, cultural practices, and chemical inter-

ventions, supplemented by data-driven decision sup-

port systems. By integrating mobile applications,

deep learning, and machine learning into existing

IPM frameworks, researchers aim to enhance the efﬁ-

ciency, precision, and ecological sustainability of pest

management efforts.

3 METHODOLOGY

The methodology employed in this study aimed to de-

velop and implement an automated detection and pre-

diction system for Lissachatina fulica infestations in

the Gal

apagos Archipelago.

3.1 In-situ Collection and Perimeter

Delimitation for Pest Control

The pest surveillance and management process be-

gins with users reporting snail sightings both in-

side and outside the Biodiversity and Management

Area (ABG). Teams then commence the collection

Artiﬁcial Intelligence-Based Detection and Prediction of Giant African Snail (Lissachatina Fulica) Infestation in the Galápagos Islands

405

of snails, starting with perimeter delineation or prop-

erty cleaning, followed by manual collection during

the day shift. Weeds are removed, and the area is

made debris-free to aid in snail detection. Poison is

dispersed to control snail spread before the shift ends.

During the night shift, teams focus on reducing the

snail population without live sample collection. They

patrol perimeters and interiors, marking snail sight-

ings on maps and collecting them in jars for later anal-

ysis. Snails are categorized by age and vitality, then

sent to the laboratory as deceased samples. Data from

the day’s activities, including photos and coordinates,

are recorded and organized for analysis.

The ﬁnal step involves verifying snail eradication

through repeated inspections with trained dogs. This

thorough process ensures the complete removal of

snail infestations from the property. Additionally, a

requirement analysis was conducted with input from

ﬁeld personnel, management, and administrative staff

to design a system tailored to their needs, including

daily report sheets.

In the requirement gathering stage, ﬁeld personnel

identiﬁed the need for a database to store detailed in-

formation for making informed decisions. Key data

ﬁelds for each collection of the giant African snail

were speciﬁed:

• Date and time are mandatory ﬁelds, automatically

ﬁlled or selected.

• Latitude and longitude are automatically ﬁlled us-

ing the phone’s GPS.

• Comments, picture, and substrate are optional

ﬁelds.

Following this, the system design phase focused

on conceptualizing and architecting the solution com-

ponents. This involved designing a mobile applica-

tion interface for efﬁcient data collection and devel-

oping a cloud-based server infrastructure for storage

and processing. Additionally, algorithms for predict-

ing giant African snail movement patterns were de-

veloped during this phase.

3.2 Flow of Data Extraction from the

Application

The development of the mobile application in-

volved utilizing cross-platform development frame-

works such as Flutter or React Native to ensure com-

patibility with both Android and iOS platforms. Key

features including real-time data collection, GPS lo-

cation tracking, image capture, and ofﬂine data stor-

age were integrated into the application. Iterative user

testing and feedback sessions were conducted to re-

Figure 1: The application collects the tracking, which is

stored locally so that when a pest is found, the tracking

ﬁeld in the form can be ﬁlled out. When a pest is found

(L. fulica), it is mandatory to ﬁll in the information on the

context in which it was found. You can also add photos of

the pest. The form data is stored in the Google Firebase

cloud to ﬁnally have a report stored in the cloud.

ﬁne the application’s design and functionality based

on user preferences and usability considerations.

3.3 Dataset

Due to the recent development of the application and

the limited availability of data for model training, our

team had to ﬁnd alternative methods to gather data.

This led us to utilize an online platform that offers

geographical coordinates for shared routes. These

coordinates, which include latitude and longitude,

are employed to map routes on interactive interfaces

and provide accurate information about notable points

along the route. Users who contribute routes on Wik-

iloc can mark their paths using the platform’s interac-

tive map and identify signiﬁcant points along the jour-

ney. These points are recorded alongside their cor-

responding geographical coordinates, enabling other

users to precisely follow the route using GPS devices

or online mapping tools. In our quest to focus on

our speciﬁc area of interest, namely the Galapagos

Islands, we followed patterns to ensure that the rela-

tionships between coordinates were meaningful. We

collected data spanning ﬁve months and consolidated

it, adjusting ﬁeld titles and retaining only the essential

data ﬁelds required for training our model. This pro-

cess resulted in a dataset containing 56,000 entries.

3.4 Application of the K-Means

Clustering Model

After conducting the exploratory analysis of the data,

they proceeded to ﬁnd the best k or the number of

clusters by analyzing the graph, where they selected

k=3. This was chosen because, upon analyzing the

data using the elbow method, a clear inﬂection point

was observed at 3, indicating that it is the optimal

number for creating the clusters. The ﬁnal result of

the k-means algorithm is a partition of the data into k

clusters, where each cluster is represented by its cen-

ICSOFT 2024 - 19th International Conference on Software Technologies

406

troid. In this case, there are 3 clusters: the ﬁrst one

with 13,770 geographical positions, the second one

with 27,232 geographical positions, and the third one

with 15,009 geographical positions. In our case, we

selected the ﬁrst cluster of these clusters for the anal-

ysis of the results.

3.5 Predictions with Transformer Time

Series

Transformer time series models predict L. fulica

movement patterns using historical data and environ-

mental variables. The Transformer algorithm oper-

ates as follows:

1. Tokenization of Time Series: The time series

data is converted into a sequence of tokens.

2. Embedding Tokens: These tokens are embedded

into a higher-dimensional space to capture com-

plex data relationships.

3. Transformer Blocks Application: Transformer

blocks process the token sequence, capturing

long-term dependencies in the data.

4. Future Value Prediction: The output of the last

transformer block is used to predict future values,

typically through a linear layer.

The pseudo-code of the transformer is as simple

as the following:

3.5.1 Time Series Analysis and Model

Deployment

In addition to the implementation of the Transformer

algorithm, additional activities were performed to en-

sure the effectiveness and successful deployment of

the predictive model. This involved the use of time

series analysis techniques to incorporate time depen-

dencies and seasonality into the predictive model.

Next, integration and deployment activities were

carried out to implement the developed model into a

cohesive system. This included integration of the mo-

bile application with cloud-based server infrastructure

to enable seamless data transmission and storage. Im-

plementation of computer vision and machine learn-

ing models on the server enabled real-time processing

of data collected from the mobile application. Exten-

sive end-to-end testing was performed to ensure the

reliability, scalability, and security of the integrated

system.

3.5.2 Performance Evaluation and Validation

The ﬁnal phase of this research involved thorough

evaluation and validation studies to gauge the effec-

tiveness and practicality of the implemented system

Require: Input lookback time series X ∈ R

T ×N

; input

Length T ; predicted length S; variates number N; token

dimension D; Transformer block number L.

1. X = X

▷X ∈ R

N×T

2. ▷ Multi-layer Perceptron works on the last dimension to

embed series into variate tokens.

3. H

= MLP(X)

▷H

∈ R

N×D

4. for l in {1, ··· , L}:

• ▷ Run through Transformer blocks.

• ▷ Self-attention layer is applied on variate tokens.

l−1

= LayerNorm



l−1

+ Self-Attn



l−1



▷H

l−1

∈ R

N×D

• ▷ Feed-forward network is utilized for series

representations, broadcasting to each token.

= LayerNorm



l−1

+ Feed-Forward



l−1



▷H

∈ R

N×D

• ▷ LayerNorm is adopted on series representations to

reduce variates discrepancies.

5. End for

Y = MLP





▷ Project tokens back to predicted

series,

Y ∈ R

N×S

(Liu et al., 2023)

Algorithm 1: Transformer Architecture.

in real-world scenarios. Through rigorous trials and

validation exercises, feedback was gathered from end-

users and stakeholders to identify areas for improve-

ment and assess the system’s impact on key pest man-

agement outcomes, including detection efﬁciency, re-

sponse time, and resource allocation.

Controlled experiments and carefully constructed

case studies were conducted to compare the perfor-

mance of the automated detection and prediction sys-

tem with alternative methods. These efforts provided

further validation, conﬁrming the usefulness and ef-

fectiveness of the developed solution through empiri-

cal examination and comparative analysis.

4 RESULTS AND DISCUSSION

The implementation of the automated detection and

prediction system yielded promising results in ef-

fectively addressing L. fulica infestations in the

Gal

apagos Archipelago.

Artiﬁcial Intelligence-Based Detection and Prediction of Giant African Snail (Lissachatina Fulica) Infestation in the Galápagos Islands

407

4.1 Mobile Application Performance

The developed mobile application demonstrated

strong performance in facilitating real-time data col-

lection and transmission from ﬁeld staff to cloud-

based servers . Field testing indicated that the app’s

easy-to-use interface and intuitive design signiﬁcantly

improved the efﬁciency and accuracy of L. fulica

surveillance efforts. Additionally, the addition of fea-

tures such as GPS location tracking and ofﬂine data

storage ensured seamless operation in remote and

resource-constrained environments. Here we have

two of the main tabs of the User Experience of the ap-

plication. The incorporation of these user-centric de-

sign elements contributes to the application’s success

in optimizing data collection processes and enhancing

surveillance efforts.

4.2 Machine Learning Model

As previously noted, a clustering approach was em-

ployed to initialize the training process for the trans-

formers. Please refer to Fig. 2 for visual representa-

tion.

Figure 2: This image represents the different clusters given

the position of the pests found. They are the key to im-

proving the number of correct positions, as they prevent the

transformer model from better modeling the phenomenon

in a speciﬁc area.

The machine learning models deployed for pre-

dicting L. fulica movement patterns demonstrated

considerable predictive accuracy and robustness.

Time-series analysis techniques effectively captured

temporal dependencies and seasonal variations in L.

fulica populations, enabling the generation of reliable

predictive models. These models provided valuable

insights for proactive pest management strategies, al-

lowing stakeholders to anticipate and preemptively

respond to potential L. fulica outbreaks. The model

produced these predictions with remarkable accuracy

as is shown in Fig 3.; the error rate in longitude (x)

and latitude (y) inside the geographic system was

only 0.001. Every spot on the picture represents a

place where the model predicted there would be pests,

which is quite similar to what was observed. This

high degree of accuracy highlights how well the pre-

dictive model predicts the existence of pests, offer-

ing insightful information for preventive pest manage-

ment plans and facilitating prompt responses to stop

possible outbreaks.

Figure 3: This image represents 112 predicted points where

plagues were indeed found in the ﬁrst days of the 16th

month, with an error of 0.001 in the geographic system in

both x (longitude) and y (latitude).

4.3 Real and Prediction Positions with

Transformer Model

For a better understanding of the effectiveness of the

employed model, Fig. 4 illustrates the relationship

between actual x coordinates (X-axis) and predicted

coordinates using a Transformer model (Y-axis). Ac-

tual coordinates are depicted as blue dots, while the

blue line represents the trend predicted by the model.

It is evident that the model accurately predicts the ac-

tual x coordinates, as most blue dots cluster closely

around the blue line, indicating precise model predic-

tions. However, some points deviate from the blue

line, suggesting that the model predictions have a few

problems in all cases, indicating room for improve-

ment through parameter optimization.

Figure 4: Relationship between the actual X coordinates

and the coordinates predicted with our model.

ICSOFT 2024 - 19th International Conference on Software Technologies

408

Similarly, Fig. 5 illustrates the relationship be-

tween actual y-values and predicted y-values. The

model performs better for y-values compared to x-

values, indicating its stronger predictive capability for

this coordinate. However, discrepancies between ac-

tual and predicted values still exist, suggesting the

need for adjustments or modiﬁcations to enhance the

model’s accuracy, particularly for x-values. With

Figure 5: Relationship between the actual y coordinates and

the coordinates predicted with our model.

a precision of 0.001 in geographic coordinates, the

study distinguished between correctly predicted (1)

and incorrectly predicted (0) x and y coordinates in

the Fig 6. Notably, the predominant distribution

within class 1 suggests the accuracy of both the train-

ing process and the resulting outcomes. This distri-

bution underscores the effectiveness of the model in

accurately predicting geographic coordinates, thereby

instilling conﬁdence in its overall performance and re-

liability.

Figure 6: This bar chart represents the total positions from

month 16 to month 24 which are the test positions in graph 1

are the correct positions and 0 are the incorrect coordinates

which were predicted with a value of 0.001 in geographic

coordinates.

4.4 Integration and Deployment

The integration of mobile application, computer vi-

sion, and machine learning components into a co-

hesive system facilitated seamless data transmission,

processing, and storage. End-to-end testing con-

ﬁrmed the reliability, scalability, and security of the

Table 1: Metrics obtained from the three methods.

RMSE

Model X Y

Prophet 0.6734 0.1393

LSTM 0.0599 0.0121

Transformer 0.0118 0.0111

integrated system, ensuring its suitability for opera-

tional deployment in real-world scenarios. Addition-

ally, user feedback and validation studies highlighted

the system’s practical utility and ease of adoption by

ﬁeld personnel and stakeholders.

4.5 Comparison with Traditional

Methods

Comparing the automated detection system with tra-

ditional methods showed signiﬁcant improvements

in efﬁciency, response time, and resource allocation.

The automated system excelled in accuracy, speed,

and scalability, mitigating the impact of L. fulica in-

festations in the Gal

apagos Archipelago.

The section analyzes LSTM, Prophet, and Trans-

former models with 56,000 data points. While

Prophet is specialized for time series with strong sea-

sonal patterns, LSTM and Transformer models are

more versatile.

Results show that the proposed model consistently

outperforms others, as shown in Table 1.

4.6 Discussion

The results demonstrate the effectiveness of integrat-

ing mobile technology, computer vision, and machine

learning into pest management practices to achieve

more sustainable and efﬁcient outcomes. By leverag-

ing real-time data collection, automated image anal-

ysis, and predictive modeling capabilities, the devel-

oped system empowers stakeholders with timely and

actionable insights for proactive pest control strate-

gies. Moreover, the scalability and adaptability of the

system enable its potential application in other eco-

logical settings facing similar pest management chal-

lenges.

In predictions of destinations, the team uses root

mean square error (RMSE).

RMSE =

∑

i=1

− ˆy

)

(3)

Where:

• n is the total number of observations.

• y

are the observed values.

• ˆy

are the predicted values.

Artiﬁcial Intelligence-Based Detection and Prediction of Giant African Snail (Lissachatina Fulica) Infestation in the Galápagos Islands

409

5 CONCLUSIONS

In conclusion, this study introduces an innovative so-

lution for detecting and predicting L. fulica infes-

tations in the Gal

apagos Archipelago using mobile

app technology and AI. By combining real-time data

collection, automated image analysis, and predictive

modeling, our system offers a scalable and efﬁcient

approach to pest management in sensitive ecosys-

tems. This research demonstrates improved surveil-

lance efﬁciency and accuracy, enabling rapid report-

ing and proactive response to potential outbreaks.

This integration of mobile tech and AI provides ac-

tionable insights for targeted control measures, con-

tributing to the preservation of the Gal

apagos’ eco-

logical integrity.

6 FUTURE WORK

Moving forward, further research is necessary to re-

ﬁne the solution, validate predictive models, and inte-

grate additional data sources. Ongoing collaboration

with local stakeholders is crucial for successful im-

plementation and sustainability. Additionally, com-

puter vision technology will enhance efﬁciency by ac-

curately classifying L. fulica specimens from ﬁeld im-

ages, reducing manual workload and expediting data

collection.

REFERENCES

Abideen, Z. U., Sun, H., Yang, Z., Ahmad, R. Z., Iftekhar,

A., and Ali, A. (2021). Deep wide spatial-temporal

based transformer networks modeling for the next des-

tination according to the taxi driver behavior predic-

tion. Applied Sciences, 11(1):17.

Awuor, F., Otanga, S., Kimeli, V., Rambim, D., and Abuya,

T. (2019). E-pest surveillance: Large scale crop pest

surveillance and control.

Budha, P. B. and Naggs, F. (2008). The giant african land

snail lissachatina fulica ( bowdich ) in nepal. The

Malacologist, 50.

Collins, K., Keith, I., and Dawson, T. P. (2019). Countering

the threat of invasive species to the galapagos marine

reserve.

Correoso, M. (2006). Estrategia preliminar para evaluar y

erradicar achatina fulica gastroda: Achatineaceae) en

ecuador. E-RevSerZoologica, 2:6.

Cowie, R. H. (1998). Patterns of introduction of non-

indigenous non-marine snails and slugs in the hawai-

ian islands. Biodiversity and Conservation, 7.

Elias, S. A. (2022). Conservation Status of Gal

apagos En-

demic Naesiotus Land Snails, volume 1-3, pages 422–

435. Elsevier.

Gerlach, J., Barker, G. M., Bick, C. S., Bouchet, P., Brodie,

G., Christensen, C. C., Collins, T., Coote, T., Cowie,

R. H., Fiedler, G. C., Grifﬁths, O. L., Florens, F. B.,

Hayes, K. A., Kim, J., Meyer, J. Y., Meyer, W. M.,

Richling, I., Slapcinsky, J. D., Winsor, L., and Yeung,

N. W. (2021). Negative impacts of invasive predators

used as biological control agents against the pest snail

lissachatina fulica: the snail euglandina ‘rosea’ and

the ﬂatworm platydemus manokwari.

Heaton, J. (2024). Transformer timeseries. GitHub.

Khatun, K. (2018). Land use management in the galapagos:

A preliminary study on reducing the impacts of inva-

sive plant species through sustainable agriculture and

payment for ecosystem services. Land Degradation

and Development, 29.

Liu, Y., Hu, T., Zhang, H., Wu, H., Wang, S., Ma, L., and

Long, M. (2023). itransformer: Inverted transformers

are effective for time series forecasting. arXiv preprint

arXiv:2310.06625.

McCallum, B. D., Geddes, C. M., Chatterton, S., Peng, G.,

Carisse, O., Turkington, T. K., Olfert, O., Leeson, J.,

Sharpe, S., Stephens, E., Hervet, V., Aboukhaddour,

R., and Vankosky, M. (2021). We stand on guard for

thee: A brief history of pest surveillance on the cana-

dian prairies.

Miquel, S. E. and Herrera, H. W. (2014). Catalogue

of terrestrial gastropods from gal

apagos (except bu-

limulidae and succineidae) with description of a new

species of ambrosiella odhner (achatinellidae) (mol-

lusca: Gastropoda). Archiv fur Molluskenkunde, 143.

Simberloff, D. and Rejmanek, M. (2020). 100 of the World’s

Worst Invasive Alien Species: A Selection From The

Global Invasive Species Database.

Vanegas, F., Bratanov, D., Powell, K., Weiss, J., and Gon-

zalez, F. (2018). A novel methodology for improv-

ing plant pest surveillance in vineyards and crops us-

ing uav-based hyperspectral and spatial data. Sensors

(Switzerland), 18.

Warner, K. (2019). Pest Control, pages 1953–1960.

Springer Netherlands, Dordrecht.

ICSOFT 2024 - 19th International Conference on Software Technologies

410