Comparative Analysis of Machine Learning Algorithms in Fish Survival Prediction

• Published in: 2024 IEEE Bangalore Humanitarian Technology Conference (B-HTC) pp. 28 - 33
• Main Authors: Shirwaikar, Rudresh D., D'Souza, Lyzandra, Bhangle, Anirudh Santosh, Joshi, Sarang Deepak, Jathar, Nupur Ajit, Alvares, Jonathan Elroy
• Format: Conference Proceeding
• Language: English
• Published: IEEE 22.03.2024
• Subjects: Accuracy; Algorithmic tuning; Classification algorithms; Logistic regression; Machine learning; Machine learning algorithms; Multilabel classification; Prediction algorithms; Random Forest; Support vector machines; Task analysis
• DOI: 10.1109/B-HTC60740.2024.10564052

This research paper conducts an exploration of prevalent machine learning classification algorithms, emphasizing the refinement of Random Forest models to optimize performance. The initial phase involves a comparative analysis of Random Forest, k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Logistic Regression, and Naive Bayes Classification algorithms across the dataset, resulting in Random forest to be best performed among all with average accuracy of 61%. Subsequently, then we focus on enhancement of the Random Forest algorithm with random train-test split dataset, resulting in an enhanced average accuracy of 65% which is about 4% improvement to the previous accuracy and choosing the best traintest split dataset, along with fine tuning the model, improved the average accuracy to 73% which is about 12% improvement to previous accuracy which is significantly higher showcasing the significance of precise algorithmic tuning. The completion of our investigation lies in the creation of an integrated model which incorporates multiple Random Forest classifier models, each specifically trained for binary class classification, along with a comprehensive model designed for multiclass classification. This combined model excels with a peak accuracy of 82.75% for the best split, demonstrating its strong performance in classification tasks. This research not only contributes to the theoretical understanding of algorithm performance and optimization but also offers practical insights for the application of machine learning in challenging classification scenarios.