Abstract

In recent years, the prediction of start-up success has been an intriguing and difficult research topic. However, most of the research in this field has only used numerical variables from funding data to develop models, leaving out much of the important textual data from the wide range of social media sites. According to research, sentiments from social media activities can be used in improving future predictions.

Specifically, I examined a set of novel features to construct a flow within the machine learning framework. In improving previous studies further advances were made by adding new features such as Twitter sentimental and profile data. I combined financial information from CrunchBase, one of the largest public databases for start-ups, with profile and tweet data from Twitter, one of the top and largest social media databases, to thoroughly examine, analyse and predict start-up status. A total of 2,613,123 tweets (the largest-scaled sentiment data ever recorded in the literature) from over 40,000 start-ups Twitter profiles were analysed using eight specific machine learning algorithms which were all classifiers. Using Random Forest Classifier on the dataset, a True Positive Rate (TPR) of 95.5% and a False Positive Rate of 3.83% was achieved, which is the highest recorded with this data. The author’s target of achieving a low false negative was also attained where the false negative is 7.6%. The novel features, which emphasized the impact of social media presence/online legitimacy of a start-up, have proved important to the overall performance of the model by being among the most important features required in the final model showing the crucial role this type of data plays in predicting start-up's status.

In the end, I was able to demonstrate that a start-up's status can be precisely predicted using a trained machine learning model with online legitimacy as a gauge of social acceptance based on tweets and profile data from Twitter and in combination with the start-up’s funding data. In addition to attempting to predict start-up statuses, this study also contributes to the continuing debate about the significance of establishing credibility online and explains how machine learning approaches might be used in research in this area.

Predicting start-up status using funding and sentimental data