Page 86 - eBook_Proceedings of the International Conference on Digital Manufacturing V1

Proceedings of the International Conference on Digital Manufacturing –
Volume 1

RESULTS AND DISCUSSION

In this research study, the machine learning model
implementation, as shown in Figure 30, was carried out using
Google Colab, a cloud-based platform that facilitates
collaborative development of machine learning models without
the need for local setup. Essential Python libraries, such as
TensorFlow, Pandas, and Scikit-learn were employed to build and
train the models. The process began with loading the dataset into
the Colab environment, followed by importing the necessary
libraries. Data preprocessing steps, including cleaning and
transformation, were undertaken to prepare the data for model
training. Subsequently, code was written to train the model, fine-
tune its parameters, evaluate its performance, and generate visual
representations of the results. Multiple iterations were performed
to enhance the model's accuracy and robustness.

XG Boost 95% Accuracy
Unilever
Random Forest 18% Accuracy
MLModel
Implementation
ABC Industry Random Forest 98% Accuracy

Figure 30: ML Model Accuracy Results

On the first dataset, we initially used XG boost algorithm to
predict the number of items sold (Demand). It produced
significant error on first iteration. After refining the model
multiple times by changing the input features to find the best
relation between input and output variables, the model was able
have an accuracy score of 0.95. We also tried applying random
forest algorithm on the first dataset, but it was not feasible to be
applied as it generated significant error on every iteration. On the
second dataset of Industry, we applied random forest algorithm to
predict demand. This model resulted in a much-reduced error than
previous model. Multiple iterations were performed to refine the

70