Predictive Modelling (linear Regression): Agricultural Production

Load Libraries

library(tidyverse)
library(tidymodels)
library(data.table)
library(DT)
library(MASS)

df <- read.csv("C:/Users/PC/Documents/R_4DS/NUMERIC_/AgricultureData/AgrcultureDataset.csv")

glimpse(df)

Data Inspection and Cleaning

null_vars <- (sapply(df, function(x) sum(is.na(x))))
t(data.frame(null_vars)) # No missing data

## Remove Trailing White space 
df$Season <- (sapply(df$Season, function(x) trimws(x)))

## scales Area
df$Scaled_Area <- scale(df$Area)

Exploratory Data Analysis

df_num <- df %>% 
  select_if(is.numeric) %>% 
  subset()

par(mfrow= c(3,3))

invisible(lapply(names(df_num), function(col_name) 
  truehist(df_num[,col_name], main = paste("Histogram of ", col_name), xlab = NA)))

## 
library(superml)

lbl = LabelEncoder$new()
ml_df$Location = lbl$fit_transform(ml_df$Location)

## 
# library(mltools)
# library(data.table)
# 
# ml_df <- ml_df %>% 
#   mutate(Crop = as.factor(Crop))
# 
# ml_df <- data.table(ml_df)
# 
# newdata <- one_hot(ml_df)

## Train-Test
n_split <- round(0.8 * nrow(ml_df))

train_indices <- sample(1:nrow(ml_df), n_split)
train_set <- df[train_indices, ]
test_set <- df[-train_indices, ]

## Checking for Colinearity with new features
library(corrgram)
# corrgram(newdata, lower.panel=panel.shade, upper.panel=panel.cor)

Model Fitting

## Simple Linear Regression (+ it handles the Categorical Variables by default, -takes time to resolve)
mdl <- lm(Production ~ ., data = train_set)

summary(mdl)

Model Evaluation

## Checkingn for Residuals
modelResiduals <- as.data.frame(residuals(mdl)) %>% 
  ggplot(aes(residuals(mdl))) +
  geom_histogram(fill='deepskyblue', color='black')

preds <- predict(model, test_set)

## Compare
modelEval <- cbind(test_set$Production, preds)
colnames(modelEval) <- c('Actual', 'Predicted')
modelEval <- as.data.frame(modelEval)

mse <- mean((modelEval$Actual - modelEval$Predicted))
rmse <- sqrt(mse)