Dual DNN/CNN Classifiers

I’m putting together two self-contained machine-learning pipelines and would like a specialist to handle every step from raw data through to actionable results.

1. Binary DNN for network traffic
The goal is to flag traffic as malicious or safe. I have access to common sources such as PCAP dumps, flow records and log exports, and can provide whichever format you prefer. You’ll decide on the most suitable feature-extraction strategy, build the deep-neural network in TensorFlow/Keras (or PyTorch if you have a strong preference), train it, and document precision, recall and ROC-AUC. Please add at least five sample predictions with explanations and finish with a concise set of ideas on how we could push performance further (e.g., alternative architectures, data augmentation, ensemble methods).

2. CNN for image categorisation
Here the task is to separate animals from vehicles. The image set contains a mix of photography and stylised renders; expect varied resolutions and lighting. Your job is to prepare the dataset (resize, normalise, augment), design an image-wise CNN—transfer learning with a backbone such as ResNet or EfficientNet is fine—train, evaluate (accuracy, confusion matrix) and supply example inferences. As above, close with a short optimisation roadmap.

Deliverables
• Clean, well-commented Python notebooks or scripts for each classifier
• Saved model weights and the full preprocessing pipeline
• Evaluation report (PDF or Markdown) covering metrics, sample predictions and improvement suggestions
• Quick-start guide so I can reproduce results on my own machine
2000words report (1000 each for both system

I’m comfortable sharing additional data and can label more samples if that will noticeably boost accuracy. Let me know your preferred framework, any hardware assumptions and the timeline you foresee.