1: Define Your Search Scope

Before you open Google Scholar, write down 3–5 keywords related to your topic. If your project is on “crop disease detection using deep learning,” your keywords might be:

• crop disease detection
• plant disease classification
• convolutional neural network agriculture
• image segmentation plant leaf

You’ll combine these with databases. Keep your scope tight — don’t try to survey all of machine learning.

2: Find the Right Papers

Use these sources. In order of preference for engineering/CS:

• IEEE Xplore — best for electronics, CS, communications
• Google Scholar — broad, good for finding citing papers
• Scopus / Web of Science — good for checking if a journal is indexed
• ScienceDirect — strong for life sciences and engineering
• ACM Digital Library — software, HCI, computing

Aim for papers from the last 5–7 years unless you’re referencing a foundational method. If you’re citing a 2008 paper, there should be a good reason.

Set a realistic target: 15–25 papers for a project report, 40–80+ for a standalone survey paper.

3: Screen and Select

Not every paper you find belongs in your survey. Apply a quick filter:

Include if:

• The paper directly relates to your problem, method, or domain
• It’s published in a peer-reviewed journal or conference (check: is it Scopus-indexed? IEEE? Springer?)
• It’s cited reasonably well (Google Scholar citation count is a rough proxy for quality)

Exclude if:

• It’s a workshop abstract or non-peer-reviewed blog post
• The methodology is completely unrelated to your work
• You can’t access the full text and can only read the abstract

4: Read and Take Structured Notes

This is where most students go wrong. They skim the abstract, copy the title, and move on. That produces a shallow survey that your guide can spot in five seconds.

For each paper, note:

• Problem they solved (one sentence)
• Method/technique used
• Dataset used (if applicable)
• Results / accuracy reported
• Limitation they mentioned (this is gold — limitations point to gaps)

A simple spreadsheet works fine. By the time you’ve done this for 20 papers, the patterns become obvious.

5: Identify Themes and Group Papers

Don’t just present papers one by one. Group them by approach, technique, or problem type.

For example, if you’re writing a survey on fraud detection:

• Group 1: Rule-based methods
• Group 2: Machine learning approaches (SVM, Random Forest)
• Group 3: Deep learning approaches (LSTM, Autoencoders)
• Group 4: Graph-based methods

This structure makes your survey readable and shows analytical thinking — not just a list.

6: Write the Survey

Each group gets its own paragraph or subsection. A well-written survey paragraph looks something like this:

Several studies have applied convolutional neural networks (CNNs) for automated plant disease detection. Mohanty et al. [1] trained a CNN on the PlantVillage dataset and reported 99.35% accuracy under controlled conditions, though the model showed significantly lower performance on field images. Building on this, Ferentinos [2] proposed a deeper architecture using AlexNet and GoogLeNet variants, achieving improved robustness across multiple crop species. However, both approaches relied on high-quality, close-up images, which limits real-world applicability in low-resource agricultural settings.

Notice what that paragraph does:

• Cites specific papers with findings, not just “many researchers have studied this”
• Connects one study to the next logically
• Ends with a limitation — setting up your proposed work

Write every paragraph like this. It’s not hard once you have your notes.

7: State the Research Gap

After your thematic paragraphs, write 2–3 sentences explicitly stating what’s still missing. This is the most important part — it’s the justification for your entire project.

Example:

Despite significant progress in automated disease detection, existing methods largely depend on curated laboratory datasets and controlled lighting conditions. There is a lack of robust, lightweight models capable of real-time classification on low-cost devices in outdoor agricultural environments. The proposed work addresses this gap by…

8: Format Your References

Use the citation style your institution requires. For most engineering institutions in India, IEEE format is standard.

IEEE format looks like: [1] M. S. Nixon and A. S. Aguado, Feature Extraction and Image Processing, 2nd ed. Oxford, UK: Academic Press, 2008.

Use Mendeley or Zotero (both free) to manage your references. Don’t format them manually — you will make mistakes and waste time.