MSc Project: Machine Learning (ML)
Student name: Sriskantharaja Mithushan
Course: MSc in Data Science and Analytics
Project title: A Comparative Evaluation of Machine Learning Techniques for Sales Forecasting
Rationale: What was the reason/motivation for choosing the project?
The motivation behind choosing this project stemmed from the increasing importance of accurate sales forecasting in business decision-making. Companies rely heavily on predictive models to optimize inventory management, plan marketing strategies, and drive revenue growth. I was particularly interested in how different machine learning techniques could enhance the accuracy of these predictions, compared to traditional forecasting methods. My goal was to explore and compare the effectiveness of various machine learning models in improving sales forecasts, which could have a significant impact on business operations and profitability.
Machine learning (ML) is a branch of artificial intelligence (AI) and computer science that focuses on using data and algorithms to enable machines to learn from experience, much like humans do. In analyzing data, the ML algorithm, processes data multiple times to learn / adjust itself to improve accuracy.
Q. Brief overview of the practical implementation?
The project was conducted in several stages, starting with data gathering, followed by preprocessing, model execution, and evaluation. A rich dataset was collected from various retail and e-commerce platforms, containing sales records, customer demographics, product categories, and revenue figures. This data was cleaned and transformed to handle missing values, normalize scales, and ensure proper formatting for model training.
Five machine learning models were implemented:
- Random Forest
- Support Vector Regression (SVR)
- LightGBM
- XGBoost
- Gated Recurrent Unit (GRU) Neural Network
Each model was trained on the prepared dataset using Python, with libraries such as Scikit-learn, LightGBM, XGBoost, and TensorFlow.
After training, predictions from each model were compared to the actual sales data. Visualization tools like Matplotlib and Seaborn were used to graphically depict the performance of each model, with side-by-side comparisons of RMSE and MAPE metrics. These visualizations helped to highlight the strengths and weaknesses of the various models.
The Profit Over Time graph below, tracks monthly profit trends, providing a clear view of how profitability fluctuates over time. Key insights include: Seasonality and Profit Growth or Decline. For example Seasonality highlights periods of increased or decreased profits, often aligned with sales cycles or specific marketing efforts. This helps in identifying high-profit months and adjusting strategies for low-profit periods.
This visualization is crucial for understanding financial performance, aiding in strategic decision-making, and optimizing resource allocation for long-term profitability.
This pairplot chart below, simultaneously shows the distributions (diagonal plots) and relationships (scatter plots) between the key variables: Sales, Quantity, Discount, and Profit. For example the scatter shows a strong positive relationship, confirming that higher sales lead to greater profits. A weak negative trend suggests that offering larger discounts may slightly lower profits. This pair plot provides a comprehensive overview of how these variables relate to each other and how each is distributed, helping in identifying trends, correlations, and potential outliers.
The chart below is a Heat-map. The Heatmap shows the correlations between the key business metrics of Sales, Quantity, Discount, and Profit. The color intensity represents the strength and direction of the relationships, with darker colors indicating stronger correlations. For example, a deep hue between Sales and Profit highlights that as sales increase, profits rise significantly. This visual tool helps identify how these variables interact and guide strategic decisions on pricing, sales, and profit optimization.
Q. Overview of outcomes/conclusions?
The project concluded with Random Forest emerging as the top-performing model in terms of prediction accuracy.
The research delved into the use of machine learning techniques for sales forecasting in the retail and e-commerce sectors, with the goal of identifying which models provide the most accurate predictions. The study examined five machine learning algorithms: Random Forest, Support Vector Regression (SVR), LightGBM, XGBoost, and Gated Recurrent Unit (GRU) neural networks. The models were evaluated using metrics such as Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE).
This study contributes to the understanding of how machine learning can be applied to sales forecasting in retail and e-commerce. It shows that tree-based ensemble methods, particularly Random Forest, are among the most effective techniques. However, deep learning models like GRU also show potential, particularly for capturing temporal dependencies. A balanced approach, combining multiple models and fine-tuning hyperparameters, can lead to more accurate sales predictions. By acting on these insights, retail and e-commerce companies can improve their forecasting accuracy, optimize inventory management, and ultimately enhance customer satisfaction and profitability.
Q. Please share some top tips/advice for students?
Completing a Bachelor of Science (BSc) or a Master of Science (MSc) at a university like the University of Wales Trinity Saint David (UWTSD) can be a rewarding and challenging journey.
The University offers a variety of Computing courses. Ensure you choose a program that aligns with your interests and career goals. During both BSc and MSc, you’ll often have the flexibility to choose elective modules. Select modules that allow you to develop key skills that are in-demand in your field, or that attract your personal interest. Balancing lectures, labs, independent study, and personal commitments is crucial. Use digital tools to organize deadlines, assignment dates, and exam preparation to stay on track. The University also offer career services to help students prepare for employment. Take advantage of these CV workshops, interview practice, and employability training.
For further information about Computing courses at UWTSD, please click-here.
~
UWTSD Applied Computing Blog 