NUS SDS DATATHON (Category A) - Tomato Cats

Inspiration

Our inspiration for this project stemmed from the desire to explore the predictive capabilities of various machine learning models in the domain of sales prediction. We were intrigued by the challenge of leveraging different algorithms to forecast 'Sales (Domestic Ultimate Total USD)' for companies based on the provided dataset.

What it does

Our project aims to demonstrate the effectiveness of different machine learning models in predicting sales for various companies. By utilizing techniques such as XGBoost, CatBoost, and GradientBoosting, we seek to uncover patterns and relationships within the data that can help improve sales forecasting accuracy.

How we built it

Our team embarked on the project by carefully cleaning and analyzing the dataset and identifying key features relevant to sales prediction. We then employed a combination of data preprocessing techniques and feature engineering to prepare the data for modeling. Using libraries like XGBoost, CatBoost, and scikit-learn, we built and trained multiple machine learning models tailored to the task at hand.

Challenges we ran into

Throughout the project, we encountered several challenges that tested our problem-solving skills and creativity:

Handling missing values: Dealing with missing data required us to devise strategies such as imputation techniques or leveraging natural language processing to predict possible values and classify datapoints.
Categorizing companies into different industries: Determining the industry of each company involved careful analysis and classification based on available information.
Converting categorical variables into indicator variables: Transforming categorical data into a format suitable for machine learning models posed its own set of challenges, requiring us to encode categorical variables in an effective manner.
Model selection: Choosing the most appropriate machine learning algorithms and tuning hyperparameters proved to be crucial for achieving optimal performance.

Accomplishments that we're proud of

We take pride in the knowledge and skills gained throughout the project, particularly in the following areas:

Handling missing values: We explored various imputation approaches, including advanced techniques like NLP, and learned valuable lessons from each method used.
Data cleaning and visualization: Through extensive data preprocessing and visualization, we gained insights into the relationships and patterns within the dataset.
Understanding different ML models: By implementing and comparing multiple machine learning models, we deepened our understanding of their respective strengths and weaknesses in the context of sales prediction.

What we learned

Our journey with this project has been enriching and educational. Some of the key takeaways include:

Data cleaning: We acquired hands-on experience in cleaning and preparing real-world datasets for machine learning tasks.
Data visualization: Visualizing data helped us identify trends, outliers, and relationships that informed our modeling decisions.
Machine learning models: We gained proficiency in implementing and evaluating various machine learning algorithms, honing our skills in model selection and performance optimization.
Use different Python packages to solve real-world problems.

What's next for NUS DATATHON (Category A) - Sales Prediction

Looking ahead, our project lays the foundation for several exciting future endeavors:

Enhanced Model Tuning: We plan to further refine our machine learning models by conducting extensive hyperparameter tuning. This involves fine-tuning the configuration of algorithms to maximize predictive accuracy and generalizability.
Industry-Specific Customization: Recognizing the diverse nature of industries, we intend to customize our models for specific sectors. This will involve tailoring feature engineering and model parameters to align with the unique characteristics of different business domains. In summary, the journey doesn't end here. NUS DATATHON (Category A) - Sales Prediction is a stepping stone towards a dynamic and evolving project, driven by the pursuit of innovation and excellence in the realm of sales forecasting through machine learning.