Fine-Tuning EfficientNetB3 for Binary Sentiment Classification: Insights into Happiness and Sadness

Recent breakthroughs in Neural networks and deep learning have tremendously enhanced sentiment analysis and provided more accurate assessments of human emotions. With the EfficientNetB3 model-known for its harmony between model size and accuracy-this work seeks to classify human emotions into “happy” and “sad” categories. While conventional machine learning approaches find it difficult to grasp the nuances of human mood, convolutional neural networks (CNNs) such EfficientNetB3 excel at processing data including complex spatial hierarchies. We want to increase sentiment-categorizing accuracy by applying the scalability and efficiency of the model. Drawn from the open-source Kaggle platform, the dataset consists of 469 “sad” and 878 “happy” images all individually categorized to ensure quality. After pre-processing including resizing and augmentation, we modified EfficientNetB3 to manage this binary classification challenge. Our method showed good and fast ability to separate the two emotions with a 93% classification accuracy. Through better emotional understanding in numerous contexts, this work highlights EfficientNetB3's potential in extending sentiment analysis, therefore supporting the more general objectives of increasing mental well-being, enhancing social inclusion, and promoting sustainable development.