For thousands of years, farming and animal husbandry have been at the heart of human concerns. It began 10,000 years ago with subsistence farming. Then, little by little, a production and trade agriculture was created.
Today, information about markets and their organisation, techniques and know-how benefiting from advances in agronomy, high-tech products, instruments and methods developed by the agri-supply industries, are available to the farmer to obtain levels of production never before achieved in human history.
However, the margins achieved by agricultural companies in developed countries remain highly variable, depending on fluctuating sales prices and on whether or not the States provide aid, while in other countries the situation of many farmers remains precarious. On the other hand, these recent industrial developments are leading some consumers in rich countries to question the quality of food, its safety and the consequences of modern methods on the environment.
Even more worrying, recent geopolitical events and the climate crisis have shifted the main issue from the quality of food grown to the quantity available."
-
Worldwide Wheat Production (Univariate): based only on past datapoints, try to project futur values"
-
USA Wheat Yield (Multivariate): using external features (such as population data & climate change data), try to forecast future values of the time-series."
Time-series Forecasting using different methods:
- Univariate method:
- ARIMA Model (Statistical analysis)
- Random Forest (Machine Learning)
- Linear Regression (Machine Learning)
- Multivariate method:
- Random Forest (Machine Learning)
- Linear Regression (Machine Learning)
- Streamlit app: install
requirements.txtand thenstreamlit run app.py - Jupyter Notebook: https://github.com/nicoboou/time-series-forecasting/blob/main/notebook/Crops%20Yield%20Notebook.ipynb
As part of a student project, we were given a white card to :
- find a reliable and complete dataset
- clean this dataset and do the necessary data preprocessing to be able to do machine learning
- do data mining to find potentially interesting insights for the future
- use our knowledge of machine learning to perform an accurate analysis
- highlighting the results obtained with Streamlit
Crop statistics for 173 products in Africa, America, Asia, Europe and Oceania, collected from 1961 to 2019.
Data from the Food and Agriculture Organization of the United Nations (FAO)
Achieving food security for all is at the heart of FAO's efforts - ensuring that people have regular access to enough quality food to lead active and healthy lives. Our three main objectives are: the eradication of hunger, food insecurity and malnutrition; the eradication of poverty and the promotion of economic and social progress for all; and the sustainable management and use of natural resources, including land, water, air, climate and genetic resources, for the benefit of present and future generations.
Primary crops, fibre crops. Crop statistics are recorded for 173 commodities, covering the following categories: Primary crops, Primary fibre crops, Cereals, Secondary cereals, Citrus, Fruit, Jute and related fibres, Oilcake equivalent, Primary oilseeds, Dry vegetables, Roots and tubers, Green fruits and vegetables and Melons. Data are expressed in terms of area harvested, quantity produced, yield and quantity of seed. The aim is to provide comprehensive coverage of production of all primary crops for all countries and regions of the world.
Source : Organisation des Nations Unies pour l'alimentation et l'agriculture (FAO)
You will find the following structure in the .ipynb file:
- Data preprocessing
- Exploratory Data Analysis
- Time-series Analysis & Forecasting
- Univariate method
- Multivariate method
The file requirements.txtwill allow you to create your dedicated conda envwithout any missing dependency.
Just type conda create --name <env_name> --file requirements.txt in your terminal