Abstract
The issue of business failure is a recurrent problem for businesses because of market competition, technology advancement, changing demands of customers and the rising impact of social media. Traditional approaches towards predicting business failure are largely based on financial parameters and fail to incorporate important behavioral information available through customer feedback and social media interactions. This paper presents an innovative approach to predicting business failure by incorporating social media sentiment analysis with the help of the Random Forest algorithm. Data for customer opinions were collected from a publicly available Kaggle Product Review Sentiment Analysis dataset. The data was cleaned, tokenized, stop words were removed, lemmatization was performed and Term Frequency – Inverse Document Frequency (TF-IDF) features were extracted from the data. This was followed by classification of the data into three classes of sentiment namely, positive, neutral and negative and then used for training of the Random Forest algorithm in an 80:20 ratio. The developed model makes use of the customer sentiment factors in combination with business engagement measures to categorize businesses as low, medium, and high risk. Experimental evaluation results showed that the proposed model provided a prediction accuracy of 94%, a macro-average and weighted-average precision, recall, and F1-score of 0.94, while the ROC analysis revealed an AUC of 0.95, thus proving high predictive performance. The feature importance analysis also revealed that negative sentiment and number of complaints are the top factors that contribute towards predicting the business failure. Results show that the application of sentiment intelligence along with machine learning in business failure prediction outperforms the traditional approach based on financial data.