You are on page 1of 5

Exercise 4: Implementing Multiple Linear Regression

Model using Neural Networks


John Vincent M. Cabugnason
College of Computer Studies and Engineering
Jose Rizal University
Cainta Rizal
johnvincent.cabugnason@my.jru.edu

Abstract applicability[1]. Understanding deep learning requires


knowledge of the expressive capabilities of neural networks.
In this paper, we explore the combination of multiple In [2], Deep learning allows computational models that are
linear regression and neural networks to create a more composed of multiple processing layers to learn
accurate and robust model for predicting the value of a representations of data with multiple levels of abstraction
dependent variable based on multiple independent variables. [2].
We discuss the limitations of multiple linear regression and
how neural networks can help capture non-linear In [3], Multiple linear regression (MLR) is a
relationships between variables. We review related statistical technique used to model the relationship between
literature, including studies that use these methods to predict a dependent variable and multiple independent variables.
cancer patient survival, student academic performance, and The problem it aims to solve is how to predict or explain the
energy consumption in residential buildings. By combining variation in the dependent variable based on the values of
these two approaches, we can create a more powerful and the independent variables. According to home.csulb.edu,
accurate model that can better represent the complexities of many difficulties tend to arise when there are more than five
real-world data. This has important implications for fields independent variables in a multiple regression equation. The
like finance, healthcare, and marketing, where accurate issue of two or more independent variables having a high
predictions are crucial for decision-making and outcomes. degree of correlation with one another is one of the most
prevalent. Multicollinearity is the term for this [3].

I. INTRODUCTION The motivation for enhancing MLR with neural


networks is to improve the accuracy and predictive power of
Multiple linear regression is a statistical method the model. MLR has limits when it comes to comprehending
that can be used to predict the value of a dependent variable complex data or capturing non-linear relationships, while
based on the values of multiple independent variables. being an effective technique for modeling relationships
Neural Network is the Machine Learning that can used to between variables. By combining MLR with neural
learn complex relationships between data. By combining networks, we can create a more powerful model that can
multiple linear regression and neural networks, it is possible better capture the nuances and complexities of real-world
to create a model that is more accurate and robust than data.
either method alone. II. REVIEW OF RELATED LITERATURE
Formula and Calculation of Multiple Linear Regression A. Multiple Linear Regression and Neural Networks
Yi=β0+β1xi1+β2xi2+...+βpxip+ϵ in Regression Tasks
where, for i = n observations:
yi=dependent variable Comparison of Artificial Neural Networks and Logistic
xi=explanatory variables Regression for 30-days Survival Prediction of Cancer
β0=y-intercept (constant term) Patients
βp=slope coefficients for each explanatory variable
ϵ=the odel’s error term (also known as the residuals)
According to Dogu et al. (2020), A machine
learning technique that imitates neural system and brain can
In [1], Machine learning has become an essential
provide better than traditional methods like logistic
tool for extracting regularities in the data and for making
regression for survival prediction and create an algorithm by
inferences. Neural networks, in particular, give the
determining influential factors [4]. In recent years, the
flexibility and scalability required to transform complex
increasing number of cancer patients and long-term
datasets into well-structured models with broad
oncological treatments have increased emergency admission
rates in our everyday practice, due to both pharmacological
side effects and clinical symptoms caused by cancer itself.

In [4], Accurate 30-days survival predictions,


directing patients to palliative care at the right time will
preventboth patients and caregivers from suffering and
ensure correct treatments. Survival analyzes should be
performed as accurately as possible. Carefully studying the
impact of patient data, including clinical, laboratory, KPS,
and PPS outcomes, on survival estimation is necessary.

According to several studies, patients with


resectable lung cancer who have high CRP levels had a
worse prognosis. Besides, the CRP level was associated
with reduced serum albumin, resulting in progressive weight Figure 1. Pseudocode of the Cuckoo Search (CS).
loss, poor performance, the higher mortality rate in cancer
patients In [5], Cuckoo search is an optimization algorithm
Predicting Student Academic Performance: A introduced by Yang and Deb. This algorithm was inspired
Comparison of Two Meta-Heuristic Algorithms Inspired by the special lifestyle of the cuckoo species. The aim of the
by Cuckoo Birds for Training Neural Networks CS is to maximize the survival rate of the eggs. Each egg in
a nest represents a solution, and a cuckoo egg stands for a
According to Chen et al. (2014), Predicting student new solution. In this case, the host bird can either throw the
academic performance has long been an important research egg away, or abandon the nest to build a new one in a new
topic. Among the issues of higher education institutions, place.
questions concerning admissions remain important [5]. The Based on the above-mentioned rules, the steps of the CS can
primary goal of the admissions process is to identify be described as the pseudocode in Figure 1. The algorithm
applicants who will be successful if they are admitted to the can be extended when each nest has multiple eggs
university. High accuracy academic performance predictions representing a set of solutions.
help with admissions decisions and improve educational
services at educational institutions.
Comparison of multi linear regression and artificial
In [5], Different approaches have been applied to neural network to predict the energy consumption of
predicting student academic performance, including residential buildings
traditional mathematical models and modern data mining
techniques [5]. These methods described the quantitative According to Nainwal and Sharman 2023, The
relationships between the inputs and outputs using a series building sector along with its sub-sectors is the world's
of mathematical formulas. As a result, it becomes necessary leading consumer of all forms of energy and consumes
to put out a model that forecasts student success based on about 30% of the world's final energy consumption. With
the outcomes of standardized tests, such as those used for the increase in population, industrialization and urbanization
college admissions and high school diploma examinations, trend, the energy sector has grown rapidly in the past
as well as other important variables. [5] The prediction is decades to cater to the increase in energy demand. To cater
accurate if the error between the predicted and actual values this, many regulatory efforts in the form of energy
is within a small range [5]. efficiency and conservation codes offering guidelines and
measures during and post design have been proposed in
different countries of the world.

Regression is a tool that is frequently used to


explain how dependent factors transform into independent
variables that are inputs into the instrument. According to
Zhao et. al., Regression models gained popularity in the
field of research as a means of addressing the three
problems. Firstly, for predicting the energy consumption
based on climatic parameters. Secondly, for predicting
energy indexes. Thirdly,
for determining the important attributes of energy
consumption, which helps in understanding the thermal V. ANALYSIS
behavior of the building.
Hyper-parameters play a crucial role in determining the
In [6], In the recent years, Artificial Neural performance of a model. You can observe and analyze the
Network (ANN) is the most widely used method for impact of hyper-parameters on your model's performance by
predicting the building‟s energy consumption as it is very changing them and observing how the model behaves.
good at dealing with the non-linear problems.
Some of the hyper-parameters that you can experiment
with include the learning rate, batch size, number of epochs,
number of layers, number of hidden units, activation
functions, and regularization techniques. By varying these
hyper-parameters, you can determine which values work
best for your dataset and model architecture.

Building and training neural networks can be

Figure 2. Energy consumption profile by different


categories of appliances.

It is seen in Figure 2 that the maximum energy is


used by the space heating appliances and minimum energy
is used by the appliances for ventilation

III. METHODOLOGY computationally intensive, and if you have limited


resources, it may be challenging to experiment with
different hyper-parameters or large datasets.

To overcome these challenges, it's important to


carefully plan the experiments, allocate sufficient resources,
IV. RESULT and monitor the model's performance. You can also consider
using automated tools or techniques such as hyper-
parameter tuning or transfer learning to streamline the
process and improve the results.

VI. CONCLUSION
Neural networks are a type of tool that can help us VII. REFERENCES
understand and analyze complex data. They are designed to 1. Montavon, G. (2020). Introduction to
work with data that is not always straightforward or easy to Neural Networks. In Machine Learning Meets
understand, and they can capture relationships between Quantum Physics (pp. 37–62). Springer
different variables that are not always linear or International Publishing.
straightforward. In other words, they can help us make sense
of data that might be too complicated for other tools to
handle. By combining multiple linear regression with neural 2. LeCun, Y., Bengio, Y., & Hinton, G. (2015).
networks, we can create a more powerful model that can Deep learning. Nature, 521(7553), 436–444.
better capture the nuances and complexities of real-world https://doi.org/10.1038/nature14539
data.
3. (N.d.). Csulb.edu. Retrieved October 3,
While multiple linear regression is a powerful tool 2023, from
for analyzing data, it has some limitations when it comes to https://home.csulb.edu/~msaintg/ppa696/696r
handling complex data or relationships that are not egmx.htm#:~:text=Many%20difficulties
straightforward. In other words, it may not be able to fully %20tend%20to%20arise,This%20is%20called
capture the nuances of real-world data. This is where neural %20multicollinearity.
networks come in - they are designed to handle more
complex data and can capture non-linear relationships 4. Arkin, F., Aras, G., & Dogu, E. (2020).
between variables. By combining these two approaches, we Comparison of artificial neural networks and
can create a more powerful and accurate model that can logistic regression for 30-days survival
better represent the complexities of the real world. prediction of cancer patients. Acta Informatica
Medica: AIM: Journal of the Society for
Combining multiple linear regression with neural Medical Informatics of Bosnia &
networks is a powerful approach to analyzing complex data Herzegovina: Casopis Drustva Za Medicinsku
and capturing non-linear relationships between variables. Informatiku BiH, 28(2), 108.
This can have important implications for fields like finance, https://doi.org/10.5455/aim.2020.28.108-113
healthcare, and marketing, where accurate predictions are
crucial for decision-making and outcomes. By using these 5. Chen, J.-F., Hsieh, H.-N., & Do, Q. (2014).
tools together, we can create a more advanced model that Predicting student academic performance: A
helps us make more informed decisions based on a more comparison of two meta-heuristic algorithms
complete picture of the data. inspired by cuckoo birds for training neural
networks. Algorithms, 7(4), 538–553.
https://doi.org/10.3390/a7040538

6. Nainwal, R., & Sharma, A. (2022).


Comparison of multi linear regression and
artificial neural network to predict the energy
consumption of residential buildings. IOP
Conference Series. Earth and Environmental
Science, 1084(1), 012005.
https://doi.org/10.1088/1755-
1315/1084/1/012005

You might also like