Kennedy Senagi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Kennedy Senagi has a total of 82 co-authors. Here are the top co-authors by number of shared publications:
- Henri Edouard Zefack Tonnang
University of KwaZulu-Natal
4 shared publications Bonface Onyango
Data Management, Modeling and Geo‐Information (DMMG) Unit International Centre of Insect Physiology and Ecology Nairobi Kenya
2 shared publicationsCaleb Kibet
Data Management, Modeling and Geo‐Information (DMMG) Unit International Centre of Insect Physiology and Ecology Nairobi Kenya
2 shared publicationsChrysantus M. Tanga
Affiliation Data Management, Modelling and Geo-Information Unit, International Centre of Insect Physiology and Ecology, Kenya
2 shared publications- Chrysantus Mbi Tanga
International Centre of Insect Physiology and Ecology , Nairobi
2 shared publications Daniel Masiga
International Centre of Insect Physiology and Ecology, Nairobi, Kenya
2 shared publications
Latest Publications
- Open Access
- 9 October 2026
- Published by Springer Nature in Communications Medicine
Sub-national estimates of visceral leishmaniasis and optimisation of geographical accessibility to treatment centres in Kenya
Background: Visceral leishmaniasis (VL) remains a major public health challenge in Kenya, disproportionately affecting populations with limited access to timely diagnosis and treatment. National case totals mask local heterogeneity in risk and access. Here we estimate sub-national VL incidence, identify its determinants and optimise geographical placement of treatment centres. Methods: We analysed nine years (October 2016-September 2025) of ward-level VL surveillance data from health facilities, with WorldPop population denominators. Bayesian small-area models related case occurrence and counts to environmental, socioeconomic and accessibility covariates. Travel time to the 46 existing treatment centres was computed over a motorised friction surface, and additional sites prioritised to maximise population coverage. Results: We record 11,985 cases, an estimated incidence of 10 per 100,000 population per year, with marked seasonality and epidemic periods. Burden is highly focal: five northern counties account for 86% of cases, with further clustering at sub-county and ward levels and predicted incidence exceeding 100 per 100,000 person-years in several wards in Turkana and West Pokot. Higher temperature, poverty, and poor housing are associated with greater probability of occurrence and higher case counts; higher precipitation is inversely associated. Only 45% of the population in need can reach a treatment centre within 120 minutes of motorised travel. Five optimally sited additions increase this 120-minute coverage to 76%, with diminishing gains from further expansion. Conclusion: Integrating small-area disease modelling with geospatial accessibility analysis reveals hidden inequities in VL burden and supports equitable, efficient deployment of treatment services towards elimination.
- 9 October 2026
- Published by Springer Nature in Communications Medicine
Open AccessSub-national estimates of visceral leishmaniasis and optimisation of geographical accessibility to treatment centres in Kenya
Background: Visceral leishmaniasis (VL) remains a major public health challenge in Kenya, disproportionately affecting populations with limited access to timely diagnosis and treatment. National case totals mask local heterogeneity in risk and access. Here we estimate sub-national VL incidence, identify its determinants and optimise geographical placement of treatment centres. Methods: We analysed nine years (October 2016-September 2025) of ward-level VL surveillance data from health facilities, with WorldPop population denominators. Bayesian small-area models related case occurrence and counts to environmental, socioeconomic and accessibility covariates. Travel time to the 46 existing treatment centres was computed over a motorised friction surface, and additional sites prioritised to maximise population coverage. Results: We record 11,985 cases, an estimated incidence of 10 per 100,000 population per year, with marked seasonality and epidemic periods. Burden is highly focal: five northern counties account for 86% of cases, with further clustering at sub-county and ward levels and predicted incidence exceeding 100 per 100,000 person-years in several wards in Turkana and West Pokot. Higher temperature, poverty, and poor housing are associated with greater probability of occurrence and higher case counts; higher precipitation is inversely associated. Only 45% of the population in need can reach a treatment centre within 120 minutes of motorised travel. Five optimally sited additions increase this 120-minute coverage to 76%, with diminishing gains from further expansion. Conclusion: Integrating small-area disease modelling with geospatial accessibility analysis reveals hidden inequities in VL burden and supports equitable, efficient deployment of treatment services towards elimination.
- Open Access
- 30 January 2026
- Published by Science Publishing Group in American Journal of Networks and Communications
- Vol. 15 (1) , 10-26
Predicting Customer Churn in the Telecommunications Industry using Machine Learning Techniques
Voluntary customer churn constitutes a persistent financial risk for telecommunications operators, particularly within enterprise customer segments where high-value accounts administer complex, multi-subscription portfolios. Industry data indicate that acquiring a new account costs between five and seven times more than retaining an existing one. Despite heightened industry awareness, the majority of operational retention platforms remain reactive, detecting departure only after the event has occurred. This investigation constructs and evaluates a machine learning pipeline engineered to identify enterprise customer churn risk proactively, drawing on authentic operational records extracted from a business-tobusiness telecommunications environment. The study follows the Cross-Industry Standard Process for Data Mining (CRISP-DM) lifecycle. A dataset of 8,454 unique business accounts, characterised by 14 raw attributes and enriched to a final 22-variable feature set, underpins the empirical work. Pronounced class imbalance, churned accounts representing approximately 6.5minority ratio of 14.3:1, necessitated specialised resampling prior to classifier training. Five oversampling strategies were benchmarked; SVMSMOTE produced the largest gain in minority-class sensitivity and was adopted for all subsequent training cycles. Ten classifier families were trained and assessed, including EasyEnsembleClassifier, RUSBoostClassifier, XGBoost, LightGBM, CatBoost, Histogram Gradient Boosting, Balanced Bagging, a multilayer perceptron, a soft-voting ensemble, and a stacking ensemble. EasyEnsembleClassifier emerged as the leading model, attaining an F1-score of 0.129 and a recall of 38.242 of 110 churned accounts. Post-hoc explainability analysis through SHAP and LIME identified active subscriber rate, geographic billing zone, and engineered interaction terms as the dominant predictive signals. The framework was operationalised within a FastAPI-based application supporting realtime individual scoring, batch CSV prediction, and retention campaign monitoring. The projected annual revenue protection under conservative assumptions exceeds 74,000 currency units. The study illustrates that interpretable, explainability-augmented machine learning frameworks can bridge the gap between quantitative model output and managerial action, offering a replicable blueprint for data-driven churn governance in both emerging and mature telecommunications markets.
- 30 January 2026
- Published by Science Publishing Group in American Journal of Networks and Communications
- Vol. 15 (1) , 10-26
Open AccessPredicting Customer Churn in the Telecommunications Industry using Machine Learning Techniques
Voluntary customer churn constitutes a persistent financial risk for telecommunications operators, particularly within enterprise customer segments where high-value accounts administer complex, multi-subscription portfolios. Industry data indicate that acquiring a new account costs between five and seven times more than retaining an existing one. Despite heightened industry awareness, the majority of operational retention platforms remain reactive, detecting departure only after the event has occurred. This investigation constructs and evaluates a machine learning pipeline engineered to identify enterprise customer churn risk proactively, drawing on authentic operational records extracted from a business-tobusiness telecommunications environment. The study follows the Cross-Industry Standard Process for Data Mining (CRISP-DM) lifecycle. A dataset of 8,454 unique business accounts, characterised by 14 raw attributes and enriched to a final 22-variable feature set, underpins the empirical work. Pronounced class imbalance, churned accounts representing approximately 6.5minority ratio of 14.3:1, necessitated specialised resampling prior to classifier training. Five oversampling strategies were benchmarked; SVMSMOTE produced the largest gain in minority-class sensitivity and was adopted for all subsequent training cycles. Ten classifier families were trained and assessed, including EasyEnsembleClassifier, RUSBoostClassifier, XGBoost, LightGBM, CatBoost, Histogram Gradient Boosting, Balanced Bagging, a multilayer perceptron, a soft-voting ensemble, and a stacking ensemble. EasyEnsembleClassifier emerged as the leading model, attaining an F1-score of 0.129 and a recall of 38.242 of 110 churned accounts. Post-hoc explainability analysis through SHAP and LIME identified active subscriber rate, geographic billing zone, and engineered interaction terms as the dominant predictive signals. The framework was operationalised within a FastAPI-based application supporting realtime individual scoring, batch CSV prediction, and retention campaign monitoring. The projected annual revenue protection under conservative assumptions exceeds 74,000 currency units. The study illustrates that interpretable, explainability-augmented machine learning frameworks can bridge the gap between quantitative model output and managerial action, offering a replicable blueprint for data-driven churn governance in both emerging and mature telecommunications markets.
- Open Access
- 30 January 2026
- Published by Elsevier in Machine Learning with Applications
- Vol. 23
Hybrid deep learning for anti-money laundering: Unsupervised detection of emerging schemes via feature fusion and explainable artificial intelligence
- 30 January 2026
- Published by Elsevier in Machine Learning with Applications
- Vol. 23
Open AccessHybrid deep learning for anti-money laundering: Unsupervised detection of emerging schemes via feature fusion and explainable artificial intelligence
- Open Access
- 10 October 2025
- Published by Frontiers Media SA in Frontiers in Digital Health
- Vol. 7, 1643314
Spatial analytics to elucidate the incubation period and drivers of visceral leishmaniasis: case of Turkana County in Kenya
Introduction: Visceral leishmaniasis (VL) is a severe and neglected tropical disease of public health concern. VL is fatal if not treated. Kenya has experienced multiple outbreaks of the disease since 2017. The underlying drivers of the disease risk dynamics, as well as the incubation period, are not well understood. Methods: We implemented statistical (spatial logistic regression and Bayesian spatial) and machine learning (random forest, support vector machine, AdaBoost, logistic regression, and extra trees) models to estimate the incubation period and predict areas of low/high risk in Turkana County, an endemic VL foci in Kenya. Two-year (2019–2020) patient data were sourced from 12 VL treatment centers in Turkana County. Environmental and weather data were sourced from satellites, while demographic data were extracted from the Kenyan Population and Housing Census 2019 dataset. The environmental and weather data were lagged up to 8 months to mimic the disease incubation period.Results: The AdaBoost was the best-performing classifier with an area under the curve of the receiver operating characteristic value of 71.2%. The model predicted three months as the optimal incubation period. Age, distance to a healthcare facility, mean monthly humidity, greenness, and total precipitation were identified as the five main predictors. The epidemiological risk map (for December 2024) was generated and deployed on the Web (https://dudumapper.icipe.org/). The Kerio Delta, Lokori, and the shores of the Lake Turkana regions were predicted to have a mid to high risk/number of cases.Discussion: These data-driven findings can improve the understanding of VL risk dynamics and support decision makers in the preparation, mitigation, and elimination of VL.
- 10 October 2025
- Published by Frontiers Media SA in Frontiers in Digital Health
- Vol. 7, 1643314
Open AccessSpatial analytics to elucidate the incubation period and drivers of visceral leishmaniasis: case of Turkana County in Kenya
Introduction: Visceral leishmaniasis (VL) is a severe and neglected tropical disease of public health concern. VL is fatal if not treated. Kenya has experienced multiple outbreaks of the disease since 2017. The underlying drivers of the disease risk dynamics, as well as the incubation period, are not well understood. Methods: We implemented statistical (spatial logistic regression and Bayesian spatial) and machine learning (random forest, support vector machine, AdaBoost, logistic regression, and extra trees) models to estimate the incubation period and predict areas of low/high risk in Turkana County, an endemic VL foci in Kenya. Two-year (2019–2020) patient data were sourced from 12 VL treatment centers in Turkana County. Environmental and weather data were sourced from satellites, while demographic data were extracted from the Kenyan Population and Housing Census 2019 dataset. The environmental and weather data were lagged up to 8 months to mimic the disease incubation period.Results: The AdaBoost was the best-performing classifier with an area under the curve of the receiver operating characteristic value of 71.2%. The model predicted three months as the optimal incubation period. Age, distance to a healthcare facility, mean monthly humidity, greenness, and total precipitation were identified as the five main predictors. The epidemiological risk map (for December 2024) was generated and deployed on the Web (https://dudumapper.icipe.org/). The Kerio Delta, Lokori, and the shores of the Lake Turkana regions were predicted to have a mid to high risk/number of cases.Discussion: These data-driven findings can improve the understanding of VL risk dynamics and support decision makers in the preparation, mitigation, and elimination of VL.
- Open Access
- 7 July 2025
- Published by Science Publishing Group in American Journal of Artificial Intelligence
- Vol. 9 (2) , 167-185
Bridging Swahili Communication Gaps: Real-Time Audio-to-Text Sentiment Analysis via Pre-trained NLP
The global proliferation of digital communication highlights a critical gap in language technologies for digitally under-represented languages, particularly Kiswahili, a language spoken by over 100 million people. While significant advancements have been made in natural language processing (NLP) for high-resource languages like English, a persistent challenge remains in creating robust computational systems for low-resource linguistic contexts. This study addresses this challenge by presenting a novel, end-to-end Kiswahili audio processing pipeline that unifies three core capabilities; real-time speech recognition, sentiment analysis, and text summarization. The system’s novelty lies in its strategic leverage of state-of-the-art, pre-trained machine learning models, including Wav2vec2, DistilBERT, and T5, demonstrating a viable approach to bridging the digital communication gap for Kiswahili in real-world applications. Our methodology involved a rigorous evaluation of the integrated system using the Mozilla Common Voice Corpus. The results revealed key insights and promising performance metrics. The speech recognition component, a foundational element of the pipeline, achieved an exceptionally low Word Error Rate (WER) of 0.3329 with the Wav2vec2 model, highlighting its capacity for accurate transcription in a low-resource setting. This is a significant finding, as it suggests that models specifically fine-tuned for such environments can overcome the challenges of data scarcity and linguistic diversity. The summarization component also demonstrated strong capabilities, yielding a ROUGE-L score of 0.6622, which indicates robust semantic and structural alignment with reference texts. While the sentiment analysis revealed a notable data imbalance with a predominance of negative samples, the model achieved a 60% accuracy, demonstrating its potential for further refinement. These findings underscore both the immense potential and the inherent limitations of applying pre-trained models to a low-resource language like Kiswahili. They provide a compelling proof of concept for the technical feasibility of Kiswahili audio processing and emphasize the critical need for continued investment in dataset expansion and model optimization. The study concludes that this work establishes a foundational groundwork for continued research and the subsequent development of advanced NLP tools specifically tailored for Kiswahili-speaking populations, ultimately aiming to improve access to education, healthcare, and information services, and to foster greater digital inclusion throughout East Africa.
- 7 July 2025
- Published by Science Publishing Group in American Journal of Artificial Intelligence
- Vol. 9 (2) , 167-185
Open AccessBridging Swahili Communication Gaps: Real-Time Audio-to-Text Sentiment Analysis via Pre-trained NLP
The global proliferation of digital communication highlights a critical gap in language technologies for digitally under-represented languages, particularly Kiswahili, a language spoken by over 100 million people. While significant advancements have been made in natural language processing (NLP) for high-resource languages like English, a persistent challenge remains in creating robust computational systems for low-resource linguistic contexts. This study addresses this challenge by presenting a novel, end-to-end Kiswahili audio processing pipeline that unifies three core capabilities; real-time speech recognition, sentiment analysis, and text summarization. The system’s novelty lies in its strategic leverage of state-of-the-art, pre-trained machine learning models, including Wav2vec2, DistilBERT, and T5, demonstrating a viable approach to bridging the digital communication gap for Kiswahili in real-world applications. Our methodology involved a rigorous evaluation of the integrated system using the Mozilla Common Voice Corpus. The results revealed key insights and promising performance metrics. The speech recognition component, a foundational element of the pipeline, achieved an exceptionally low Word Error Rate (WER) of 0.3329 with the Wav2vec2 model, highlighting its capacity for accurate transcription in a low-resource setting. This is a significant finding, as it suggests that models specifically fine-tuned for such environments can overcome the challenges of data scarcity and linguistic diversity. The summarization component also demonstrated strong capabilities, yielding a ROUGE-L score of 0.6622, which indicates robust semantic and structural alignment with reference texts. While the sentiment analysis revealed a notable data imbalance with a predominance of negative samples, the model achieved a 60% accuracy, demonstrating its potential for further refinement. These findings underscore both the immense potential and the inherent limitations of applying pre-trained models to a low-resource language like Kiswahili. They provide a compelling proof of concept for the technical feasibility of Kiswahili audio processing and emphasize the critical need for continued investment in dataset expansion and model optimization. The study concludes that this work establishes a foundational groundwork for continued research and the subsequent development of advanced NLP tools specifically tailored for Kiswahili-speaking populations, ultimately aiming to improve access to education, healthcare, and information services, and to foster greater digital inclusion throughout East Africa.
- Open Access
- 24 January 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (1) , e0292418
Mathematical and computational modeling for organic and insect frass fertilizer production: A systematic review
Organic fertilizers have been identified as a sustainable agricultural practice that can enhance productivity and reduce environmental impact. Recently, the European Union defined and accepted insect frass as an innovative and emerging organic fertilizer. In the wider domain of organic fertilizers, mathematical and computational models have been developed to optimize their production and application conditions. However, with the advancement in policies and regulations, modelling has shifted towards efficiencies in the deployment of these technologies. Therefore, this paper reviews and critically analyzes the recent developments in the mathematical and computation modeling that have promoted various organic fertilizer products including insect frass. We reviewed a total of 35 studies and discussed the methodologies, benefits, and challenges associated with the use of these models. The results show that mathematical and computational modeling can improve the efficiency and effectiveness of organic fertilizer production, leading to improved agricultural productivity and reduced environmental impact. Mathematical models such as simulation, regression, dynamics, and kinetics have been applied while computational data driven machine learning models such as random forest, support vector machines, gradient boosting, and artificial neural networks have also been applied as well. These models have been used in quantifying nutrients concentration/release, effects of nutrients in agro-production, and fertilizer treatment. This paper also discusses prospects for the use of these models, including the development of more comprehensive and accurate models and integration with emerging technologies such as Internet of Things.
- 24 January 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (1) , e0292418
Open AccessMathematical and computational modeling for organic and insect frass fertilizer production: A systematic review
Organic fertilizers have been identified as a sustainable agricultural practice that can enhance productivity and reduce environmental impact. Recently, the European Union defined and accepted insect frass as an innovative and emerging organic fertilizer. In the wider domain of organic fertilizers, mathematical and computational models have been developed to optimize their production and application conditions. However, with the advancement in policies and regulations, modelling has shifted towards efficiencies in the deployment of these technologies. Therefore, this paper reviews and critically analyzes the recent developments in the mathematical and computation modeling that have promoted various organic fertilizer products including insect frass. We reviewed a total of 35 studies and discussed the methodologies, benefits, and challenges associated with the use of these models. The results show that mathematical and computational modeling can improve the efficiency and effectiveness of organic fertilizer production, leading to improved agricultural productivity and reduced environmental impact. Mathematical models such as simulation, regression, dynamics, and kinetics have been applied while computational data driven machine learning models such as random forest, support vector machines, gradient boosting, and artificial neural networks have also been applied as well. These models have been used in quantifying nutrients concentration/release, effects of nutrients in agro-production, and fertilizer treatment. This paper also discusses prospects for the use of these models, including the development of more comprehensive and accurate models and integration with emerging technologies such as Internet of Things.
- Open Access
- 8 January 2025
- Published by Springer Nature in International Journal of Tropical Insect Science
- Vol. 45 (1) , 101-120
Automatic synthesis of insects bioacoustics using machine learning: a systematic review
The application of machine learning has received increasing attention in the synthesis of insect sounds to preserve biodiversity. This study reviewed current literature on the application of these techniques in the automatic synthesis of insect bioacoustic and their applications in insects as food and feed, improving pest management, and as well as managing pollinators. To achieve this, the study used Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology to identify, screen, and include the final articles used in this review, based on criteria such as papers addressing machine learning in insect acoustics, biodiversity, ecology conservation, etc. This study revealed that most of the researchers used secondary data and the microphone was the common tool used to record sound signals. Sound signals were mainly pre-processed using techniques such as denoising, segmentation, and windowing. Sound signal classification algorithms were categorized mainly as shallow and deep machine learning algorithms. In the shallow machine learning algorithms, the most common method of feature extraction was the Mel-Frequency Cepstral Coefficient (MFCC) and the Support Vector Machine (SVM) was the most commonly used algorithm. In deep learning, spectrogram image features were widely extracted and the Convolutional Neural Network (CNN) was mostly used to synthesize the spectral features. This paper also reviewed recent developments in insect bioacoustics signals processing, applications, and future directions. Generally, machine learning algorithms can be applied and deployed successfully to different insects’ automatic synthesis problems to improve the production of insects (as food and/or feed), and improve/preserve diversity and life on Earth.
- 8 January 2025
- Published by Springer Nature in International Journal of Tropical Insect Science
- Vol. 45 (1) , 101-120
Open AccessAutomatic synthesis of insects bioacoustics using machine learning: a systematic review
The application of machine learning has received increasing attention in the synthesis of insect sounds to preserve biodiversity. This study reviewed current literature on the application of these techniques in the automatic synthesis of insect bioacoustic and their applications in insects as food and feed, improving pest management, and as well as managing pollinators. To achieve this, the study used Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology to identify, screen, and include the final articles used in this review, based on criteria such as papers addressing machine learning in insect acoustics, biodiversity, ecology conservation, etc. This study revealed that most of the researchers used secondary data and the microphone was the common tool used to record sound signals. Sound signals were mainly pre-processed using techniques such as denoising, segmentation, and windowing. Sound signal classification algorithms were categorized mainly as shallow and deep machine learning algorithms. In the shallow machine learning algorithms, the most common method of feature extraction was the Mel-Frequency Cepstral Coefficient (MFCC) and the Support Vector Machine (SVM) was the most commonly used algorithm. In deep learning, spectrogram image features were widely extracted and the Convolutional Neural Network (CNN) was mostly used to synthesize the spectral features. This paper also reviewed recent developments in insect bioacoustics signals processing, applications, and future directions. Generally, machine learning algorithms can be applied and deployed successfully to different insects’ automatic synthesis problems to improve the production of insects (as food and/or feed), and improve/preserve diversity and life on Earth.
- Open Access
- 1 June 2024
- Published by Wiley in Ecology and Evolution
- Vol. 14 (6) , e11457
WiPFIM: A digital platform for interlinking biocollections of wild plants, fruits, associated insects, and their molecular barcodes
The current knowledge on insects feeding on fruits is limited, and some of the scarce existing data on the fruit‐associated insects are secluded within the host institutions. Consequently, their value is not fully realized. Moreover, in countries like Kenya, the integration of biocollections data within a digital framework has not been fully exploited. To address these gaps, this article presents a description of the development of a web‐based platform for data sharing and integrating biodiversity historical data of wild plants, fruits, associated insects, and their molecular barcodes (WiPFIM) while leveraging data science technologies. The barcodes corresponding to the biocollections data were retrieved from BOLD database. The platform is an online resource about fruit–insect interactions that can be of interest to a worldwide community of users and can be useful in building innovative tools. The platform is accessible online at https://test‐dmmg.icipe.org/wpfhi.
- 1 June 2024
- Published by Wiley in Ecology and Evolution
- Vol. 14 (6) , e11457
Open AccessWiPFIM: A digital platform for interlinking biocollections of wild plants, fruits, associated insects, and their molecular barcodes
The current knowledge on insects feeding on fruits is limited, and some of the scarce existing data on the fruit‐associated insects are secluded within the host institutions. Consequently, their value is not fully realized. Moreover, in countries like Kenya, the integration of biocollections data within a digital framework has not been fully exploited. To address these gaps, this article presents a description of the development of a web‐based platform for data sharing and integrating biodiversity historical data of wild plants, fruits, associated insects, and their molecular barcodes (WiPFIM) while leveraging data science technologies. The barcodes corresponding to the biocollections data were retrieved from BOLD database. The platform is an online resource about fruit–insect interactions that can be of interest to a worldwide community of users and can be useful in building innovative tools. The platform is accessible online at https://test‐dmmg.icipe.org/wpfhi.
- Open Access
- 14 May 2024
- Published by Frontiers Media SA in Frontiers in Artificial Intelligence
- Vol. 7, 1403593
A convolutional neural network with image and numerical data to improve farming of edible crickets as a source of food—A decision support system
Crickets (Gryllus bimaculatus) produce sounds as a natural means to communicate and convey various behaviors and activities, including mating, feeding, aggression, distress, and more. These vocalizations are intricately linked to prevailing environmental conditions such as temperature and humidity. By accurately monitoring, identifying, and appropriately addressing these behaviors and activities, the farming and production of crickets can be enhanced. This research implemented a decision support system that leverages machine learning (ML) algorithms to decode and classify cricket songs, along with their associated key weather variables (temperature and humidity). Videos capturing cricket behavior and weather variables were recorded. From these videos, sound signals were extracted and classified such as calling, aggression, and courtship. Numerical and image features were extracted from the sound signals and combined with the weather variables. The extracted numerical features, i.e., Mel-Frequency Cepstral Coefficients (MFCC), Linear Frequency Cepstral Coefficients, and chroma, were used to train shallow (support vector machine, k-nearest neighbors, and random forest (RF)) ML algorithms. While image features, i.e., spectrograms, were used to train different state-of-the-art deep ML models, i,e., convolutional neural network architectures (ResNet152V2, VGG16, and EfficientNetB4). In the deep ML category, ResNet152V2 had the best accuracy of 99.42%. The RF algorithm had the best accuracy of 95.63% in the shallow ML category when trained with a combination of MFCC+chroma and after feature selection. In descending order of importance, the top 6 ranked features in the RF algorithm were, namely humidity, temperature, C#, mfcc11, mfcc10, and D. From the selected features, it is notable that temperature and humidity are necessary for growth and metabolic activities in insects. Moreover, the songs produced by certain cricket species naturally align to musical tones such as C# and D as ranked by the algorithm. Using this knowledge, a decision support system was built to guide farmers about the optimal temperature and humidity ranges and interpret the songs (calling, aggression, and courtship) in relation to weather variables. With this information, farmers can put in place suitable measures such as temperature regulation, humidity control, addressing aggressors, and other relevant interventions to minimize or eliminate losses and enhance cricket production.
- 14 May 2024
- Published by Frontiers Media SA in Frontiers in Artificial Intelligence
- Vol. 7, 1403593
Open AccessA convolutional neural network with image and numerical data to improve farming of edible crickets as a source of food—A decision support system
Crickets (Gryllus bimaculatus) produce sounds as a natural means to communicate and convey various behaviors and activities, including mating, feeding, aggression, distress, and more. These vocalizations are intricately linked to prevailing environmental conditions such as temperature and humidity. By accurately monitoring, identifying, and appropriately addressing these behaviors and activities, the farming and production of crickets can be enhanced. This research implemented a decision support system that leverages machine learning (ML) algorithms to decode and classify cricket songs, along with their associated key weather variables (temperature and humidity). Videos capturing cricket behavior and weather variables were recorded. From these videos, sound signals were extracted and classified such as calling, aggression, and courtship. Numerical and image features were extracted from the sound signals and combined with the weather variables. The extracted numerical features, i.e., Mel-Frequency Cepstral Coefficients (MFCC), Linear Frequency Cepstral Coefficients, and chroma, were used to train shallow (support vector machine, k-nearest neighbors, and random forest (RF)) ML algorithms. While image features, i.e., spectrograms, were used to train different state-of-the-art deep ML models, i,e., convolutional neural network architectures (ResNet152V2, VGG16, and EfficientNetB4). In the deep ML category, ResNet152V2 had the best accuracy of 99.42%. The RF algorithm had the best accuracy of 95.63% in the shallow ML category when trained with a combination of MFCC+chroma and after feature selection. In descending order of importance, the top 6 ranked features in the RF algorithm were, namely humidity, temperature, C#, mfcc11, mfcc10, and D. From the selected features, it is notable that temperature and humidity are necessary for growth and metabolic activities in insects. Moreover, the songs produced by certain cricket species naturally align to musical tones such as C# and D as ranked by the algorithm. Using this knowledge, a decision support system was built to guide farmers about the optimal temperature and humidity ranges and interpret the songs (calling, aggression, and courtship) in relation to weather variables. With this information, farmers can put in place suitable measures such as temperature regulation, humidity control, addressing aggressors, and other relevant interventions to minimize or eliminate losses and enhance cricket production.
- Open Access
- 30 April 2024
- Published by MDPI AG in Sustainability
- Vol. 16 (9) , 3746
Unveiling Environmental Influences on Sustainable Fertilizer Production through Insect Farming
Entomocomposting is fast and environmentally friendly, boosts soil quality and crop production, and improves resilience to climate change. The black soldier fly larvae (BSFL) catalyze the composting process, but their efficiency is highly influenced by environmental factors and the quality of the substrate. This study employs response surface methodology to discern physical–chemical factors that influence the nutrient quality of BSF frass fertilizer. Internet of Things (IoT) sensors were deployed to monitor in real-time both independent variables (air temperature, moisture content, humidity, and substrate temperature) and dependent variables (nitrogen, phosphorous, and potassium); the data were relayed to the cloud. A non-linear regression model was used to study the relationship between the dependent and independent variables. Results showed that air humidity and air temperature did not have a significant effect on nitrogen and phosphorus accumulation in frass fertilizer, respectively, but phosphorus was significantly influenced by air humidity. On the other hand, neither air temperature nor moisture content has a significant effect on potassium concentration in frass fertilizer. We found that an air temperature of 30 °C and 41.5 °C, substrate temperature of 32.5 °C and 35 °C, moisture content between 70 and 80%, and relative humidity beyond 38% can be conducive for the production of high-quality BSF frass fertilizer. Model validation results showed better robustness of prediction with values of 63–77%, and values of 62–76% for nitrogen, phosphorous, and potassium. Our findings highlight the potential for the application of digital tools as a fast and cost-effective decision support system to optimize insect farming for the production of high-quality frass fertilizer for use in sustainable agriculture and crop production.- 30 April 2024
- Published by MDPI AG in Sustainability
- Vol. 16 (9) , 3746
Open AccessUnveiling Environmental Influences on Sustainable Fertilizer Production through Insect Farming
Entomocomposting is fast and environmentally friendly, boosts soil quality and crop production, and improves resilience to climate change. The black soldier fly larvae (BSFL) catalyze the composting process, but their efficiency is highly influenced by environmental factors and the quality of the substrate. This study employs response surface methodology to discern physical–chemical factors that influence the nutrient quality of BSF frass fertilizer. Internet of Things (IoT) sensors were deployed to monitor in real-time both independent variables (air temperature, moisture content, humidity, and substrate temperature) and dependent variables (nitrogen, phosphorous, and potassium); the data were relayed to the cloud. A non-linear regression model was used to study the relationship between the dependent and independent variables. Results showed that air humidity and air temperature did not have a significant effect on nitrogen and phosphorus accumulation in frass fertilizer, respectively, but phosphorus was significantly influenced by air humidity. On the other hand, neither air temperature nor moisture content has a significant effect on potassium concentration in frass fertilizer. We found that an air temperature of 30 °C and 41.5 °C, substrate temperature of 32.5 °C and 35 °C, moisture content between 70 and 80%, and relative humidity beyond 38% can be conducive for the production of high-quality BSF frass fertilizer. Model validation results showed better robustness of prediction with values of 63–77%, and values of 62–76% for nitrogen, phosphorous, and potassium. Our findings highlight the potential for the application of digital tools as a fast and cost-effective decision support system to optimize insect farming for the production of high-quality frass fertilizer for use in sustainable agriculture and crop production.