Vai al contenuto principale

Karina Brotto Rebuli

Phd thesis

Development of Machine Learning Models Applied to Milk Production of Holstein Friesian Cows in Automatic Milking Systems

This PhD project is part of the Circular Health for Industry (CH4I) project, whose general goals are (i) to study how to collect, manage and analyse data in a Circular Health approach without keeping separate data silos of human health, animals and plants and (ii) to develop Artificial Intelligence algorithms that work on them, to improve the infrastructure for collecting and analysing such data and to reskill workers for the adoption of AI technologies (https://ch4i.di.unito.it/). Within the CH4I, the work is being developed as part of the Subproject 2 Agrifood, specifically in the context of milk production of a herd that uses an Automatic Milking System (AMS) exclusively from 2016.

AMS, also called milking robots, generate a plenty of data on milk production and quality, milking behaviour and udder health, providing detailed information about each cow. This was not easily obtained with previous conventional systems and the extensive collection of data through the AMS has led to an exponentially growing amount of data [1]. However, very little has been done with this data to model milk production [2-9]. The goal of this PhD project is to develop methodologies and algorithms to fill this gap, particularly on long-term milk production modelling with Machine Learning (ML) methods.

Long-term prediction in milk production remains a challange [10]. However, the approach proposed in my PhD thesis has been achieving good results. The developed framework is based on the definition of Productivity Groups, in terms of daily average milk yield, and the analysis of the stability of these groups.

The first stage was the development of a robust clustering framework, that uses four different clustering algorithms and, then, merge their results. This approach is more robust than traditional single-algorithm clustering due the application of two components. First, the use of more than one algorithm ensures that its outcome is not biased by any algorithm pattern. Second, the merging index we proposed takes into account the quality of the results of the single algorithms to produce the final outcome. Once the Low and High Productivity Groups are defined, their continuty over lactaiton perdios is analysed.

This multi-clustering framework was first applied to a mid-size farm. Preliminar results were discussed at the International Conference on Precision Dairy Farming, Vienna 2022. The article with final results, entitled Multi-algorithm clustering analysis for characterizing cow productivity on automatic milking systems over lactation periods [11], was publisehd at the Computers and Electronics in Agriculture journal. After the success of this first use-case, it was validated in other 16 farms. Prelimiar results were presented at the European Conference of Precision Livestock farming 2024. The detailed results are under review to be published as a scientific article.

In general, the Productivity Groups are not stable - what would be expected, considering how hardship is to make long-term prediction on cow's milk production.

Another very important finding of this research was that farms that are more capable of mantaining the cows for later lactation cycles are also those in which the High Productivity Groups grow or, at least, are stable over lactation cycles.

A deeper investigation on the main differences between farms that can retain cows in production for more lactation periods (referred as continued productivity from now on) and those that cannot (referred as non-continued productivity from now on) revealed that cows in the continued productivity farms produce a richer milk in terms of protein, and are less stressed in the first lactation period.

The next steps of the research are (i) to generate an interpretable machine learning models to automatize the framework that defines to which Productivity Group a cow belongs to, and (ii) to generate an interpretable machine learning model for the herd dynamics of the Productivity Groups. Interpretable models have the potential for generating knowledge on the process that is being modelled. An Interpretable Genetic Programming model is being developed and applied on predicting the Productivity Groups of future lactation periods. The preliminary studies on this theme were presented in the poster A Comparison of Structural Complexity Metrics for Explainable Genetic Programming [12] at the Genetic and Evolutionary Computation Conference (GECCO), Lisbon 2023.

 

References:

[1] Jacobs, J. A., J. M. Siegford (2012). Invited review: The impact of automatic milking systems on dairy cow management, behavior, health, and welfare. Journal of Dairy Science, 95(5), 2227-2247.
[2] Masía, F. M., Lyons, N. A., Piccardi, M., Balzarini, M., Hovey, R. C., Garcia, S. C. (2020) Modeling variability of the lactation curves of cows in automated milking systems. Journal of Dairy Science, 103, 8189– 8196.
[3] Fuentes, S., Viejo, C. G., Cullen, B. , Tongson, E., Chauhan, S., S., Dunshea, F. R.(2020) Artificial Intelligence Applied to a Robotic Dairy Farm to Model Milk Productivity and Quality based on Cow Data and Daily Environmental Parameters. Sensors 2020, 20(10), 2975.
[4] Piwczynski, D., SitkowskaB., Aerts, J., Schork, P. M. 2020. Forecasting the milk yield of cows on farms equipped with automatic milking system with the use of decision trees. Animal Science Journal 2020, 91.
[5] Klis, P., Piwczynski, D., Sawa, A., Sitkowska, B. (2021). Prediction of Lactational Milk Yield of Cows Based on Data Recorded by AMS during the Periparturient Period. Animals 2021, 11, 383.
[6] Ji, B., Banhazi, T., Phillips, C. J.C., Wang, C., Li, B. (2022) A machine learning framework to predict the next month's daily milk yield, milk composition and milking frequency for cows in a robotic dairy farm. Biosystems Engineering, 216, 186–197.
[7] Gauthier, R.; Largouët, C'; Dourmad, JY. (2022) Prediction of litter performance in lactating sows using machine learning, for precision livestock farming, Computers and Electronics in Agriculture, 196, 2022, 106876.
[8] Streefland, GJ.; Herrema, F.; Martino, M. (2023) A Gradient Boosting model to predict the milk production, Smart Agricultural Technology, 6, 2023, 100302. 
[9] Gauthier, R.; Largouët, C'; Dourmad, JY. Prediction of litter performance in lactating sows using machine learning, for precision livestock farming, Computers and Electronics in Agriculture, 196, 2022, 106876.
[10] Bovo, M. Ozella, L., Fiorilla, E.; Forte C. (2024). Assessment of the future productivity level of dairy cows: between dream and reality. In: Proceedings of the 11th European Conference on Precision Livestock Farming, Bologna, Italy.
[11] Rebuli, K. B., Ozella, L., Vanneschi, L., Giacobini M. (2023). Multi-algorithm clustering analysis for characterizing cow productivity on automatic milking systems over lactation periods, Computers and Electronics in Agriculture, Volume 211, 2023, 108002.
[12] Rebuli, K. B., Giacobini, M., Silva, S., Vanneschi, V. (2023). A Comparison of Structural Complexity Metrics for Explainable Genetic Programming. In Proceedings of the Companion Conference on Genetic and Evolutionary Computation (GECCO '23 Companion). Association for Computing Machinery, New York, NY, USA, 539–542. 

 

Research activities

Attended Congresses

  • The 11th European Conference on Precision Livestock Farming (ECPLF 2024), 9th to 12th September, 2024, Bologna, Italy.
  • EvoStar 2024 (European Conference on Genetic Programming, International Conference on the Applications of Evolutionary Computation, European Conference on Evolutionary Computation in Combinatorial Optimization and International Conference on Computational Intelligence in Music, Sound, Art and Design), 3rd to 5th April, 2024, Aberystwyth, Wales.
  • GECCO 2023 (Genetic and Evolutionary Computation Conference), 15th to 19th July, 2023, Lisbon, Portugal. Link to the publication: https://dl.acm.org/doi/abs/10.1145/3583133.3590595
  • EvoStar 2023 (European Conference on Genetic Programming, International Conference on the Applications of Evolutionary Computation, European Conference on Evolutionary Computation in Combinatorial Optimization and International Conference on Computational Intelligence in Music, Sound, Art and Design), 11th to 15th April, 2023, Brno, Czech Republic.
  • WIVACE 2022 (XVI International Workshop on Artificial Life and Evolutionary Computation), 14th to 16th September, 2022, Gaeta, Italy. Link to the publication: https://link.springer.com/chapter/10.1007/978-3-031-31183-3_17
  • 10th European Conference on Precision Livestock Farming (ECPLF 2022) and the 3rd International Conference on Precision Dairy Farming (PDC 2022), 29th August to 2nd September, Vienna, Austria.
  • GECCO 2022 (The Genetic and Evolutionary Computation Conference), 9th to 13th July, 2022, Boston, United States. Link to the publication: https://dl.acm.org/doi/10.1145/3520304.3528806
  • EvoStar 2022 (European Conference on Genetic Programming, International Conference on the Applications of Evolutionary Computation, European Conference on Evolutionary Computation in Combinatorial Optimization and International Conference on Computational Intelligence in Music, Sound, Art and Design), 20th to 22th April, 2022, Madrid, Spain.
  • Game of Research - North Edition. 16h December, 2021. Turin, Italy.

 

Courses

  • SPECIES Summer School on Evolutionary Computation, by SPECIES Society. Moraira, Spain. 2023.
  • Trasferimento di Conoscenze, UNITO 2023.
  • Training course in Science Communication for PhD students, UNITO, 2023.
  • Summer School on Automatic Algorithm Design, by CRIStAL laboratory on the University of Lille's Cité Scientifique campus. University of Lille. 2023.
  • Science Comunication and Public Engagement, by Prof. Daniel Edward Chamberlain, Prof. Gianpiero Vigani, and Prof. Enrico Caprio. University of Torino. 2023.
  • Big Data and Digital Tools Applied to Livestock Production, by Prof. Guilherme J. M. Rosa and Prof. João Dorea. University of Padova. 2022.
  • Grant Writing: How to write a Competitive Application to a Funding Agency, by Prof. Roger Coulombe. University of Torino. 2022.
  • Bayesian Statistics for Genetics, by Prof. Ken Rice and Prof. Jonathan Wakefield. University of Washington. 2022.
  • Complex Networks: Theory, Methods, and Applications, coord. Prof. Carlo Piccardi. Lake Como School of Advanced Studies. 2022.
  • Filmaking for Scientists, by Prof. Samer Angelone. University of Torino. 2021.

 

Training periods abroad

  • Information Management School at NOVA University of Lisbon. From 6th of May 2024 to current, under the supervision of Prof. Dr. Leonardo Vanneschi.
  • Computer Science Department at Federal University of Minas Gerais, Brazil. From 29th of Spetember 2023 to 13th of March 2024, under the supervision of Prof. Dr. Gisele Pappa.
  • Information Management School at NOVA University of Lisbon. From 20th of June 2023 to 2nd
    of August 2023, under the supervision of Prof. Dr. Leonardo Vanneschi.
  • LASIGE Computer Science and Engineering Research Centre, Department of Informatics, Faculty of Sciences, University of Lisbon, Portugal. From 4th of October 2022 to 28th of February 2023, under the supervision of Prof. Dr. Sara Silva.
  • Information Management School at NOVA University of Lisbon. From 4th of April 2022 to 4th of August 2022, under the supervision of Prof. Dr. Leonardo Vanneschi.

 

PhD Programme Credits

  • Compulsory educational ECTS: 23.
  • Total other theoretical ECTS: 92.5.
  • Practical educational credits: 198.
  • Total: 313.5.

 

 

Publications

All of my research products
Last update: 19/10/2024 14:23

Location: https://dott-scivet.campusnet.unito.it/robots.html
Non cliccare qui!