Music is considered an universal language, a way to reduce language and cultural barriers between people from around the globe. The present paper proposes to dive into the subjective domain of emotion recognition, aiming to decipher the relationship between musical elements and human emotions. In the present paper, deep learning methods are leveraged to classify the mood induced by a piece of music. For feature modelling, multiple attributes are used. Spotify features are employed, such as rhythm, tone, danceability, energy, loudness, valence, speechiness, and others. Other features, such as Mel-Frequency Cepstral Coefficients, Spectral Centroid, Chroma Energy Normalized Statistics, and the Mel Spectrogram, are concatenated into an image. The conducted experiments employed deep learning methods, including a ResNet18 model, which had not been used in this context before. The best-performing model, with 86% accuracy, proves to be a hybrid one, based on ResNet18 architecture and enriched with two Bidirectional Long-Short Term Memory layers, outperforming by 6%-11% other deep learning architectures that the present paper experimented with. The results are rather similar to other studies from the literature, even though the approaches and the datasets are different.