Felipe Ferreira Bocca

The effect of tuning, feature engineering, and feature selection in data mining applied to rainfed sugarcane yield modelling

Data-mining techniques were applied to data from sugarcane production.The impact of different approaches to include weather data was evaluated.The RReliefF algorithm is used to evaluate feature engineering.We evaluated the impact of tuning, feature selection, and feature engineering in error.Sixty-six combinations were evaluated to quantify the impacts on model performance. Crop yield models can assist decision makers within any agro-industrial supply chain, even with regard to decisions that are unrelated to the crop production. Considering the characteristics of the mechanisms and data related to yield, data mining techniques are suitable candidates for modelling. The use of these techniques within a context with feature engineering, feature selection, and proper tuning can further improve performance beyond a simple replacement of multiple linear regression. To evaluate the impact of the different steps in the mentioned context, we evaluated sugarcane (Saccharum spp.) yield modelling with data obtained from a sugarcane mill. For a combination of six techniques, tuning, feature selection, and feature engineering, leading to 66 combinations, we assessed final model performance. Average performance across combinations resulted in a mean absolute error (MAE) of 6.42Mgha-1. Using different techniques led to a range of MAE from 4.57 to 8.80Mgha-1 on average. The best and worst performances for an individual model were MAEs of 4.11 and 9.00Mgha-1. Models with lower performance were close to simply predicting yield from the average yield for each number of cuts (MAE of 9.86Mgha-1). Tuning and feature engineering reduced the MAE on average by 1.17 and 0.64Mgha-1, respectively. Feature selection removed nearly 40% of the features but increased the MAE by 0.19Mgha-1. The performance of models was improved by simple strategies such as decomposing weather attributes and detailing fertilisation. Evaluation of feature importance provided by the RReliefF feature selection algorithm was used to explain the performance gains. If empirical models are needed, they will rely on using advanced techniques, but they will need proper algorithm tuning and feature engineering to extract most of the information from datasets. Based on the results, we recommend following the presented workflow for the development of yield models.

From spreadsheets to sugar content modeling

We derived empirical models for sugar content using field data from a sugarcane mill.We show how the data mining framework can be used to model sugar content.A feature selection algorithm was able to identify good predictors in the dataset.On average, models achieved errors of 2.5% of the lower value in the test set.The best models have 90% of error within 5.4kgMg1. Sugarcane mills need sugar content estimates in advance to establish their commercial strategy. To obtain these estimates, mills rely on historical averages or maturation curves. Crop models have also been developed to provide those estimates. Leveraging mill data about fields and sugar content at harvest, we developed empirical models using different data mining techniques along with the RReliefF algorithm for feature selection. The best model was attained with Random Forest with features selected by RReliefF, having a mean absolute error of 2.02kgMg1. This model outperformed Support Vector Regression and Regression Trees with and without feature selection. Models were also evaluated by the Regression Error Characteristic Curves, which showed that the best model was able to predict 90% of the observations within a precision of 5.40kgMg1.

Decision trees for knowledge discovery on the yield decline of sugarcane ratoons

Due to the high costs associated with planting a new sugarcane field, sugarcane ratooning is explored to decrease production costs. However, ratoons have successively smaller yields, because of the effect known as sugarcane yield decline, which can impair the profits. The factors underpinning the ratoon yield decline are yet to be established. The objective of this work is to apply decision trees to sugarcane production mill data to evaluate factors related to the sugarcane ratoon yield decline. For this, meteorological and production data from four sugarcane mills were evaluated, comparing the yield obtained with the yield of the following year.

Development of predictive models using Data Mining techniques to detect borer infestation (Diatraea saccharalis) in sugarcane culture

Borer infestation (Diatraea saccharalis) is one of the main concerns in the sugarcane crop because it affects productivity directly and negatively. In order to find alternatives that minimize these damages, the objective of this work is to develop predictive models using data mining tools to predict the infestation of the borer in the sugarcane crop.

Partial dependence plots for inspecting machine learning models of sugarcane yield

Sugarcane yield models are important tools for planning purposes in the sucroenergetic sector. When black-box techniques are used to create such models, methodologies such as partial dependence plots are required for further understanding them. We evaluated partial dependence plots for a few selected important variables. We observed that different techniques learned similar responses. The patterns were consistent across different techniques, feature sets, and the use of feature selection. They also reflected knowledge about the crop.

REC curves for visual evaluation of sugarcane yield machine learning models

Model validation often is performed with metrics unsuitable for the task. Also, no metric should be used alone as criterion. One alternative is the use of Regression Error Characteristic Curve (REC). The use of REC curves was able to replace the results of single metric evaluation while providing information about trade-offs and model variability. Considering the limitations of plotting several curves, REC curves should be used for final steps of model validation.