Presenter(s)
Thomas Donnelly
Abstract
This study aimed to develop a Multilayer Perceptron (MLP) that accurately predicts crop yields within 10% of ground truth in 80% of cases using weather data, region, soil type, temperature, fertilizer, irrigation, days taken to harvest, and rail fall. The dataset has 1 million unique data points and 10 columns. In order to process the data, all Boolean data had to be converted to integers, Numerical data standardized, while Categorical data was checked for non-null values. The model will be trained using a random selection of 90% of the data for training and 10% for testing. The effectiveness of the model will be derived from the accuracy and the Mean Square Error (MSE) of the model. The learning rate of 0.001 was chosen so as not to overfit the data. As of the writing of this abstract, the best model has an MSE of 0.3992, a percentage of predictions within 10% of the actual of 62.10%, and a percentage of predictions within 20% of the actual of 88.02%.
College
College of Science & Engineering
Department
Computer Science
Campus
Winona
First Advisor/Mentor
Minigrui Zhang
Second Advisor/Mentor
Sudharsan Iyengar
Third Advisor/Mentor
Trung Nguyen
Start Date
4-24-2025 9:00 AM
End Date
4-24-2025 10:00 AM
Presentation Type
Poster Session
Format of Presentation or Performance
In-Person
Session
1a=9am-10am
Poster Number
25
Included in
Using Multilayer Perceptron (MLP) to predict crop yields
This study aimed to develop a Multilayer Perceptron (MLP) that accurately predicts crop yields within 10% of ground truth in 80% of cases using weather data, region, soil type, temperature, fertilizer, irrigation, days taken to harvest, and rail fall. The dataset has 1 million unique data points and 10 columns. In order to process the data, all Boolean data had to be converted to integers, Numerical data standardized, while Categorical data was checked for non-null values. The model will be trained using a random selection of 90% of the data for training and 10% for testing. The effectiveness of the model will be derived from the accuracy and the Mean Square Error (MSE) of the model. The learning rate of 0.001 was chosen so as not to overfit the data. As of the writing of this abstract, the best model has an MSE of 0.3992, a percentage of predictions within 10% of the actual of 62.10%, and a percentage of predictions within 20% of the actual of 88.02%.