Journal of High Technology Management Research 34 (2023) 100455

Contents lists available at ScienceDirect

Journal of High Technology Management Research

journal homepage: www.elsevier.com/locate/hitech

Machine learning models for forecasting and estimation of

business operations
Shaik Fayaz Ahamed a, *, A. Vijayasankar a, M. Thenmozhi b, S. Rajendar c, P. Bindu d,
T. Subha Mastan Rao e
a
ECE Department, V R Siddhartha Engineering College, India
b
School of Social Sciences and Languages (SSL), Vellore Institute of Technology, Vellore, India
c
Electronics and Communication Engineering, Vardhaman College of Engineering, Hyderabad, India
d
Department of Mathematics, Koneru Lakshmaiah Education Foundation, Green Field, Vaddeswaram, India
e
CSE Department, CMR Technical Campus, Hyderabad, India

A R T I C L E I N F O A B S T R A C T

Keywords: Machine Learning (ML) systems are built to shift through large amounts of data. Applying ML in
Machine learning algorithms production settings allows for the collection of additional data that can be used to guide future
Forecasting decisions about the system’s design. Since the late 1970s, academics have taken an interest in the
Business operations
field of financial predictions. The real business environment has neglected statistical methods in
Sales forecasting
forecasting, despite highly sophisticated models and rising competence in econometrics and
economics studies. Current research centres on implementing various algorithms to identify the
variation in performance for each product, and it compares the time series models to one another
to identify the better model. A basic forecast model can make reliable, fact-based sales pro­
jections, as suggested by the books on forecasting. The worth of the forecast model lies in its
ability to simplify the arduous tasks of budgeting and rolling forecasting by providing an unbiased
forecast upon which a comprehensive financial strategy can be based. In this research, we first
look for appropriate machine learning algorithms that can be used to predict sales of truck
components, and then we run experiments with the selected algorithms to make predictions about
sales and assess how well they work. Business forecasting allows for the estimation of a wide
variety of activities, each of which can be tailored to the individual requirements of the company.
Here are a few examples of frequently estimated kinds of operations. Although it is well-known
that certain algorithms, such as Simple Linear Regression, Gradient Boosting Regression, Support
Vector Regression, and Random Forest Regression, outperform others, it has been demonstrated
that Random Forest Regression is the most suitable algorithm. Based on the results of the ex­
periments and the analysis, the Ridge regression algorithm was selected as the best algorithm to
conduct the sales forecasting of truck components for the selected data.

1. Introduction

The business planning and decision-making process is dynamic and poses challenges for all companies. The more accurate the
forecasts are, the better the outcomes for planning and decision-making will be in meeting the organization’s requirements (Hsieh,

* Corresponding author.
E-mail address: [email protected] (S.F. Ahamed).

https://doi.org/10.1016/j.hitech.2023.100455
Received 8 March 2023; Received in revised form 4 April 2023; Accepted 17 April 2023
Available online 8 May 2023
1047-8310/© 2023 Elsevier Inc. All rights reserved.
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

Giloni, & Hurvich, 2019). Despite the fact that management has struggled with forecasting for decades, sophisticated computer
technology now makes it possible to execute even the most complex forecasting methods with relative ease (Mutavhatsindi, Sigauke, &
Mbuvha, 2020). As the field of supply chain management expands, so does the need for an accurate planning system to help businesses
deal with fluctuating product and resource demand. The objective of any business is to stock inventory at a level sufficient to satisfy
customer demand while keeping acquisition and storage costs to a minimum (Eseye & Lehtonen, 2020; Feng, Sun, & Zhang, 2020;
Jawad et al., 2020; Mansour Saatloo, Moradzadeh, Moayyed, Mohammadpourfard, & Mohammadi-Ivatloo, 2022; Rustam et al., 2020).
Management must predict product demand in order to estimate the amount of raw materials, labour, and budget needed by the firm’s
production division. The company must plan and schedule these assets well in advance of any product demand from consumers. In
order to effectively manage stock levels at retail outlets, wholesalers, and distributors, businesses need accurate demand forecasts in
order to implement effective inventory management systems (De Caro, De Stefani, Vaccaro, & Bontempi, 2022; Ferdaus, Chakrabortty,
& Ryan, 2022; Hoque & Aljamaan, 2021; Tang, Jiao, Chen, & Gui, 2022; Tetteroo, Baratchi, & Hoos, 2022).
Companies lose money when they have an excess of inventory because they have to pay for more storage room than they need, their
stock depreciates, and perishable items go bad. Sales will drop and customer satisfaction will plummet if there isn’t enough stock to go
around. Therefore, accurate predictions are required so that the company can adjust to the new demand and continue to expand.
Results of the hypothetical prediction model and simulated sales are shown in Fig. 1. We extrapolate future sales based on the past
success of various sales entities. To simulate the constant prediction, the estimated model is a trend function raised to the sixth power
(Feng et al., 2019). In Fig. 2 we can see the difference between real sales and projections for the coming year. It is necessary to
introduce cyclicality from outside sources when the underlying econometric causal effect model is quite steady and lacks excessive
cyclicality. In this instance, the discrepancy between forecasted and realised sales is computed and displayed graphically in Fig. 2.
Fig. 2 shows that the algorithm consistently overestimates sales for Q1 and consistently underestimates sales for Q4. This is evident
from the fact that the numbers in the columns fall below 100% in Q4 and then rise above 100% in Q1 (Ahmadi et al., 2020; Ai et al.,
2023; Alkawaz, Abdellatif, Kanesan, Khairuddin, & Gheni, 2022; Chen et al., 2019; Gao, Cheng, Suganthan, & Yuen, 2022). Forecast
value is subtracted from real value to determine value. The forecast algorithm does well, though, in Q2 and Q3. Manipulating forecast
results to show more realistic estimation is possible when the forecast model provides overly smooth forecasts. The forecast inaccuracy
can be adjusted when it follows a predictable pattern. Obviously, if the business were to alter its operations, this manipulation would
no longer be effective, and the company would be better off clinging to its original forecast model. (See Fig. 3.)

2. Forecasting of business operations

The purpose of business forecasting is to predict future business conditions so that business policies can be formulated to attain
desired outcomes, and this is accomplished by analyzing past performance (Alhussein, Aurangzeb, & Haider, 2020). Data that can be
used to back up business choices is now easier to collect and store than ever before thanks to recent advances in computing and
technology. It’s safe to say that no other department has expanded as quickly as the planning department in recent years. Preliminary
in nature, business forecasting is meant to aid in the making of decisions and the laying out of plans within a company. For future
management-focused outcomes, it equips people with the ability to foresee and control key factors and variables. In layman’s terms,
business forecasting is a method for making estimates about or predictions about upcoming patterns based on information gathered
from businesses. The importance of forecasting in the modern business world has grown as companies seek to maximise profits by
minimising expenses and improving their customers’ overall happiness (Khan et al., 2020). When a man takes on the role of business
owner, he also takes on the responsibility of trying to predict the future, the success or failure of which would rely to a large degree on
his accuracy in doing so. The goal of forecasting is to lessen the degree to which future outcomes (such as costs, profits, sales, output,
prices, etc.) can’t be predicted by management. Data-driven analysis is what business forecasting is all about; it’s what allows one to
plan for and make the most of favourable future circumstances. There are many ways in which the future will unfold similarly to the
past (Srivastava et al., 2022). The following procedures are always carried out regardless of the technique employed in making the
forecast:

Fig. 1. Forecast model for a simulated sales series (estimated by the author).

2
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

Fig. 2. Forecast errors grouped to quarters

Fig. 3. Flow chart for the business forecasting operations.

Step 1: Explain why you’re using a prediction and what you hope to accomplish by doing so. Forecasting is done so that the most up-
to-date information can be used to direct future actions in the direction of the organization’s goals.
Step 2: You can choose the factors that will be used in the forecast, such as sales and exchange rates.
Step 3: To foresee shifts that are likely to follow the current level of activities, it is necessary to establish the time horizon of the
forecast, i.e. whether it is short, middle, or long term.
Step 4: Choose an appropriate forecasting model to make predictions about the future based on the causes of previous changes.
Step 5: Acquire the data you need to create the forecast.
Step 6: Predict and act on the outcome of the forecast.
By analyzing past data and looking for trends and patterns, machine learning models can be used to make predictions about how a
company will perform in the future. Time-series forecasting, regression analysis, and classification models are just a few examples of
the many machine learning methods that can be used for foresight. Predictions are made using time-series forecasting models, which
use data gathered over time to draw conclusions. Sales, revenue, and other KPIs can be predicted with these models by analyzing past
data for trends and patterns. Predictions can be made using the results of a regression analysis model, which finds relationships be­
tween factors. Sales and income projections can be made with the help of such models, which take into account variables like
advertising budgets, product prices, and market dynamics. In order to find trends and patterns in data and to extrapolate future
outcomes, classification models are employed. Customers’ actions can be predicted, and new threats and possibilities for the company
can be discovered with the help of these models. Machine learning models used for forecasting purposes need high-quality data that is
pertinent to the business processes under study in order to be effective. Remove any anomalies or mistakes from this data through
preprocessing to ensure reliable findings. To ensure the model’s predictions continue to serve the company, it must be updated and
fine-tuned over time as new data becomes available.

3
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

3. Machine learning methods for business forecasting

The field of study known as Machine Learning is focused on teaching computers how to improve themselves without being given a
set of instructions. When a computer program’s success on tasks in some class T, as measured by some performance metric P, improves
with additional experience E, we say that the programme is engaging in machine learning (Sajjad et al., 2020). Machine learning refers
to the ability of a computer programme to perform duties through the examination and analysis of data. Spam filtering, stock fore­
casting, personal assistants, product suggestions, autonomous vehicles, sentiment analysis, and other similar tasks are just some of the
common uses for machine learning. When data is collected and recorded in a systematic fashion over time, we have what is called a
time series. In the fields of economics and finance, numerous time series are regularly kept. It is a collection of numbers that are
gathered at regular periods, such as once per day, once per week, once per month, once per quarter, and once per year. The central
tenet of time series analysis is that forecasts are derived from the observed behaviour of the same variable over time. This means the
variable to be forecasted will continue to follow the same trend it has in the past. It is also believed without explicit statement that past
data can be accessed. Collecting and analyzing historical observations is a crucial step in time series forecasting. This is because it
allows researchers to create a mathematical model that accurately represents the fundamental process that generates the series’ data
(Sidogi, Mongwe, Mbuvha, & Marwala, 2022). In this work, we learn about the many categories of information that can be found in
company databases. In this volume, we also include the criteria we’ll use to judge the accuracy of the forecasting model. The algebraic
discrepancy between the actual realised value and the forecast for a given time frame constitutes the forecast error. The forecast error
for a set of forecasts should be as near to zero as feasible (Fleming & Goodbody, 2019). Based on the forecast and actual values nine
parameters are computed for each technique used in this context. These metrics allow for more precise comparisons.
The scalability of machine learning models is another perk when it comes to using them for predicting and estimation. Businesses of
all sizes and in all sectors can benefit from these models because of their ability to process both big and complex datasets and op­
erations. They can also be combined with other enterprise software to facilitate analysis and implementation of the model’s findings.
Forecasting and estimating business operations with machine learning models is innovative because of the accurate, scalable, and real-
time insights they provide, which in turn help businesses make better decisions and stay ahead of the competition.
Our primary objective is to select, fine-tune, and combine the LightGBM model with other state-of-the-art algorithms that can
forecast the YoY and QoQ earnings growth rate based on financial, macroeconomic, and market factors. Fig. 4 shows the main building
blocks of our model:

i. a dimension-reduction technique called principal component analysis (PCA),

ii. a basic machine learning classifier called LightGBM, and
iii. an optimization tool called Hyperopt to fine-tune the LightGBM hyperparameters. (See Fig. 5.)

The novelty of utilizing machine learning models for forecasting and estimating business operations lies in their ability to analyze
massive data sets and find patterns that may not be obvious to human analysts. By offering predictions and insights based on historical
data and real-time inputs, these models can help businesses make more accurate and informed choices. Additionally, as new data is
collected and processed, machine learning algorithms can learn and improve over time. This means that estimates and forecasts of
company operations can improve as more data is collected and analyzed.

4. Discussion and analysis

Fig. 6 shows that the model and the real series are not identical. The approximated model captures almost all of the dips and peaks,
with a small number of exceptions. Internal issues and the timing of the recognition of purchases can and will cause peaks and valleys.

Fig. 4. Types of machine Learning Methods for Business Forecasting.

4
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

Fig. 5. Block diagram for the Business Operations forecasting Methods for Machine learning Models.

In the residual analysis depicted at the foot of Fig. 6, discrepancies between the model and actual sales are shown as scaled residuals.
The presence of autocorrelation can be better observed and understood with the help of a visual representation. Autocorrelation can be
split into two subtypes. If autocorrelation is positive, then we can expect to see a sequence of deviations with the same sign, i.e., a
positive departure will be followed by another positive deviation, and so on. If there is a negative autocorrelation, that means that
every negative departure will be followed by a positive one. The lack of autocorrelation in the residual can be deduced using the
structure shown in Fig. 6 and the statistical test presented. (See Figs. 7–9.)
Using information gathered from a private mass manufacturing firm in Istanbul, we were able to make accurate demand pre­
dictions. Correlation analysis was first used to establish the level of dependence between the demand quantities and other factors of the
manufactured goods. Six different ML methods, including a “base state” model, were used to create a representation of the data. Given
the findings, feature engineering was applied to the data and the outcomes were contrasted. Python and the PyCharm library, both of
which are commonly used in ML methods, are employed in this investigation. Correlation analysis was applied to the dataset to
determine which variables received from the company had an impact on the demand estimation result and which did not. In order to
lower the dimensionality of features and assess their discriminatory ability in categorization models, correlation-based feature se­
lection is commonly employed. In a nutshell, this approach was used to assess the extent to which one variable affected demand and to
identify the truly independent factors at play. Daily working hours were found to be unrelated to demand in the correlation study.
Accuracy score, Mean Absolute Error, and Max Error are just some of the evaluation metrics compared in Table 1, demonstrating
that Random Forest Regression did admirably across the board. When comparing Simple Linear Regression, Gradient Boosting
Regression, and Support Vector Machine to Random Forest Regression, the latter produced the most accurate forecasts of future sales.
Across all measures, Simple Linear Regression performed the worst and had the most error. Above, we see the findings tabulated in a
condensed form.

5. Conclusion

In this research, ML algorithms were used to predict future sales for a company that uses bulk customization. Foreseeing future
needs is done with the help of the Python programming language and the PyCharm tool. Six algorithms—XGBoost, Ridge Regression,
Linear Regression, Lasso Regression, Lasso Lars, and Bayes Based Ridge Regression—have emerged as front-runners in recent years due
to their speed and accuracy in ML competitions, making them the most popular choices for demand forecasting regression analysis.
Hyperparameter optimization has been applied to these methods, so it is desirable to achieve the best possible outcome by selecting
appropriate parameters. Various parameter combinations for each algorithm have been tried and evaluated to see which ones produce
the best results. In general, business forecasting estimates what kind of activities a company will be conducting based on its own unique
requirements and objectives. Companies can make better decisions and take preventative action toward success if they have an ac­
curate picture of their upcoming operations. Although it is difficult to directly quantify and enter insider information into the
LightGBM model, analysts do benefit from a more comprehensive data set at present. While the machine learning model we are
currently using only uses structured data obtained from public databases like Compustat or Thomson Reuters, it is important to note
that the broader information set of machine learning methods with advanced techniques such as Natural Language Processing (NLP)
models may potentially exploit more non-quantitative fundamental information from various market news or messages. Our machine
learning model could be used to predict other types of earnings in the future, such as a ranking of companies in the same industry over
the same time span.

CRediT authorship contribution statement

Shaik Fayaz Ahamed: Conceptualization, Methodology, Software. A. Vijayasankar: Data curation, Writing – original draft. M.
Thenmozhi: Visualization, Investigation. S. Rajendar: Supervision. P. Bindu: Software, Validation. T. Subha Mastan Rao: Writing –
review & editing.

5
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

Fig. 6. Forecast model (fitted) and the actual sales.

Fig. 7. Plot for Means Absolute Error for different ML Algorithms.

Fig. 8. Plot for Accuracy Score for different ML Algorithms.

6
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

Fig. 9. Plot for Max Error for different ML Algorithms.

Table 1
Comparison table for evaluation of different parameters.
Algorithm Mean Absolute Error Accuracy Score (%) Max Error

Simple Linear Regression 3.28 78.19 0.52

SVM 3.26 81.5 0.49
Random Forest Regression 3.12 84.32 0.46
Gradient Boosting Regression 2.91 85.18 0.45

Data availability

No data was used for the research described in the article.

References

Ahmadi, A., Nabipour, M., Mohammadi-Ivatloo, B., Amani, A. M., Rho, S., & Piran, M. J. (2020). Long-term wind power forecasting using tree-based learning
algorithms. IEEE Access, 8, 151511–151522. https://doi.org/10.1109/ACCESS.2020.3017442
Ai, P., Xiong, C., Li, K., Song, Y., Gong, S., & Yue, Z. (2023). Effect of data characteristics inconsistency on medium and long-term runoff forecasting by machine
learning. IEEE Access, 11, 11601–11612. https://doi.org/10.1109/ACCESS.2023.3241995
Alhussein, M., Aurangzeb, K., & Haider, S. I. (2020). Hybrid CNN-LSTM model for short-term individual household load forecasting. IEEE Access, 8, 180544–180557.
https://doi.org/10.1109/ACCESS.2020.3028281
Alkawaz, A. N., Abdellatif, A., Kanesan, J., Khairuddin, A. S. M., & Gheni, H. M. (2022). Day-ahead electricity price forecasting based on hybrid regression model. IEEE
Access, 10, 108021–108033. https://doi.org/10.1109/ACCESS.2022.3213081
Chen, C., et al. (2019). CRML: A convolution regression model with machine learning for hydrology forecasting. IEEE Access, 7, 133839–133849. https://doi.org/
10.1109/ACCESS.2019.2941234
De Caro, F., De Stefani, J., Vaccaro, A., & Bontempi, G. (April 2022). DAFT-E: Feature-based multivariate and multi-step-ahead wind power forecasting. IEEE
Transactions on Sustainable Energy, 13(2), 1199–1209. https://doi.org/10.1109/TSTE.2021.3130949
Eseye, A. T., & Lehtonen, M. (Dec. 2020). Short-term forecasting of heat demand of buildings for efficient and optimal energy management based on integrated
machine learning models. IEEE Transactions on Industrial Informatics, 16(12), 7743–7755. https://doi.org/10.1109/TII.2020.2970165
Feng, C., Cui, M., Hodge, B.-M., Lu, S., Hamann, H. F., & Zhang, J. (Oct. 2019). Unsupervised clustering-based short-term solar forecasting. IEEE Transactions on
Sustainable Energy, 10(4), 2174–2185. https://doi.org/10.1109/TSTE.2018.2881531
Feng, C., Sun, M., & Zhang, J. (March 2020). Reinforced deterministic and probabilistic load forecasting via $Q$ -learning dynamic model selection. IEEE Transactions
on Smart Grid, 11(2), 1377–1386. https://doi.org/10.1109/TSG.2019.2937338
Ferdaus, M. M., Chakrabortty, R. K., & Ryan, M. J. (May 2022). Multiobjective automated type-2 parsimonious learning machine to forecast time-varying stock indices
online. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 52(5), 2874–2887. https://doi.org/10.1109/TSMC.2021.3061389
Fleming, S. W., & Goodbody, A. G. (2019). A machine learning metasystem for robust probabilistic nonlinear regression-based forecasting of seasonal water
availability in the US west. IEEE Access, 7, 119943–119964. https://doi.org/10.1109/ACCESS.2019.2936989
Gao, R., Cheng, W. X., Suganthan, P. N., & Yuen, K. F. (Oct. 2022). Inpatient discharges forecasting for Singapore hospitals by machine learning. IEEE Journal of
Biomedical and Health Informatics, 26(10), 4966–4975. https://doi.org/10.1109/JBHI.2022.3172956
Hoque, K. E., & Aljamaan, H. (2021). Impact of Hyperparameter tuning on machine learning models in stock Price forecasting. IEEE Access, 9, 163815–163830.
https://doi.org/10.1109/ACCESS.2021.3134138
Hsieh, M.-C., Giloni, A., & Hurvich, C. (Dec. 2019). The propagation and identification of ARMA demand under simple exponential smoothing: Forecasting expertise
and information sharing. IMA Journal of Management Mathematics, 31(1), 307–344. https://doi.org/10.1093/imaman/dpaa006
Jawad, M., et al. (2020). Machine learning based cost effective electricity load forecasting model using correlated meteorological parameters. IEEE Access, 8,
146847–146864. https://doi.org/10.1109/ACCESS.2020.3014086
Khan, M. A., et al. (2020). Effective demand forecasting model using business intelligence empowered with machine learning. IEEE Access, 8, 116013–116023.
https://doi.org/10.1109/ACCESS.2020.3003790
Mansour Saatloo, A., Moradzadeh, A., Moayyed, H., Mohammadpourfard, M., & Mohammadi-Ivatloo, B. (Sept. 2022). Hierarchical extreme learning machine enabled
dynamic line rating forecasting. IEEE Systems Journal, 16(3), 4664–4674. https://doi.org/10.1109/JSYST.2021.3128213
Mutavhatsindi, T., Sigauke, C., & Mbuvha, R. (2020). Forecasting hourly global horizontal solar irradiance in South Africa using machine learning models. IEEE Access,
8, 198872–198885. https://doi.org/10.1109/ACCESS.2020.3034690
Rustam, F., et al. (2020). COVID-19 future forecasting using supervised machine learning models. IEEE Access, 8, 101489–101499. https://doi.org/10.1109/
ACCESS.2020.2997311

7
S.F. Ahamed et al. Journal of High Technology Management Research 34 (2023) 100455

Sajjad, M., et al. (2020). A novel CNN-GRU-based hybrid approach for short-term residential load forecasting. IEEE Access, 8, 143759–143768. https://doi.org/
10.1109/ACCESS.2020.3009537
Sidogi, T., Mongwe, W. T., Mbuvha, R., & Marwala, T. (2022). Fusing sell-side analyst bidirectional forecasts using machine learning. IEEE Access, 10, 76966–76974.
https://doi.org/10.1109/ACCESS.2022.3193141
Srivastava, A. K., et al. (2022). Machine learning approach for forecast analysis of novel COVID-19 scenarios in India. IEEE Access, 10, 95106–95124. https://doi.org/
10.1109/ACCESS.2022.3204804
Tang, T., Jiao, D., Chen, T., & Gui, G. (2022). Medium- and long-term precipitation forecasting method based on data augmentation and machine learning algorithms.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15, 1000–1011. https://doi.org/10.1109/JSTARS.2022.3140442
Tetteroo, J., Baratchi, M., & Hoos, H. H. (2022). Automated machine learning for COVID-19 forecasting. IEEE Access, 10, 94718–94737. https://doi.org/10.1109/
ACCESS.2022.3202220