British Airways Customer Behaviour Analysis

This project is a detailed customer behaviour analysis aimed at improving the airline's service offerings and predicting booking patterns. The project encompassed sentiment analysis of customer reviews and predictive modelling to forecast booking completions, utilising various data science tools and methodologies to derive actionable insights.

Sentiment Analysis Phase

• Conducted sentiment analysis on 1,000 customer reviews of British Airways from Skytrax, an airline review platform. Utilized Python and BeautifulSoup for web scraping to gather the reviews.

• Undertook extensive data cleaning to eliminate irrelevant characters, symbols, and emojis to clean the reviews.

• Employed the NLTK library to filter out common stop words, along with specific terms related to this study such as Airline, Airport, British Airways, BA, etc.

• Used the RoBERTa model, a Natural Language Processing tool, to classify the reviews into negative, neutral, and positive sentiments.

• Mapped positive sentiments to a 1-5 rating scale for a more straightforward interpretation of results.

• Developed a Tableau dashboard to visualize sentiments, offering actionable insights into customer satisfaction and service areas needing improvement.

Predictive Modeling Phase

• Predicted customer booking completion using Machine Learning and identify the key factors influencing booking behaviour.

• Prepared and assessed the quality of the British Airways customer booking dataset. Conducted a preliminary review of the data to understand its structure and identify any immediate data quality issues.

• Performed exploratory data analysis to identify key factors affecting booking decisions, including booking origins and travel routes.

• Generated insightful visualizations using Matplotlib in Python to delineate customer preferences and booking patterns distinctly.

• Conducted correlation analysis and statistical tests to ascertain the significance of each variable.

• Transformed categorical variables through one-hot encoding.

• Developed a RandomForest classifier to predict customer booking behaviours, fine-tuned with hyperparameter optimization.

Conclusion

• A trend of negative sentiment in customer reviews, particularly centred around the customer service experience.

• Developed a RandomForest model for British Airways to predict customer booking behaviours, achieving 85% accuracy and an AUC of 0.78.

• Booking origin and route are crucial predictors for booking completion followed by purchase lead and length of stay.

• Model performance was constrained by a limited dataset and class imbalance; addressing these issues by expanding the dataset and managing class imbalance could improve predictions.

• The model serves as a strategic tool for British Airways, enhancing customer engagement and operational efficiency.