OPTIMAL WAY FOR PREDICTING START-UP COMPANY SUCCESS RATES USING MACHINE LEARNING

Abstract

Predicting the success of start-up companies has become a critical challenge, as investors often risk
significant capital without understanding potential outcomes. To address this issue, a machine learning-based
approach is proposed to predict start-up success using key parameters like funding, participants, and milestones.
The dataset used for this analysis is sourced from Kaggle’s "Startup Success Prediction" repository, which
provides valuable information for training the models. Data preprocessing steps, including handling missing
values, data imbalance correction via the ADASYN algorithm, and feature normalization, were implemented.
Multiple machine learning algorithms were employed and evaluated, such as Random Forest, KNN, SVM, Naïve
Bayes, and Logistic Regression. Among these, Random Forest achieved the highest accuracy of over 94%,
demonstrating superior performance compared to other models. The models were evaluated using metrics like
accuracy, precision, recall, F1 Score, confusion matrix, and ROC curve analysis. Data visualizations were used
to explore trends in funding, company status, and category distributions. The final model was deployed using
Flask, allowing users to input data for real-time predictions and receive actionable recommendations for
improving start-up success