Section 3: My Weekly Project Diary (The "Behind-the-Scenes")

This section documents the 10-month journey from a "Spatial Blind Spot" to a fully functional Decision Support System.

This month was about realizing that "good data" is hard to find and requires creative problem-solving.

Week 1 (Oct 1–7): The Raster Roadblock. I officially kicked off the data extraction phase, but I immediately hit a wall: the internet coverage maps I found were just "pictures" (rasters). I couldn't extract numbers for analysis.
Week 2 (Oct 8–14): The Mukim Experiment. On Oct 6, I met with Dr. Rosshairy. We decided to try a "hyper-local" approach by looking at Mukims (sub-districts). However, after a week of manual checking, I realized internet coverage was over 98% everywhere—it didn't help me distinguish "good" from "bad" areas.
Week 3 (Oct 15–21): Finding the "Early Adopters." I pivoted to Institutional Drivers. I spent the week manually mapping every University and Tertiary Hub in Malaysia. If there’s a university, there are students (the perfect e-wallet target).
Week 4 (Oct 22–31): Balancing the Scale. My supervisor pointed out that focusing only on students was too narrow. I spent this week researching how to add SME (Merchant) Density to the model to represent the business side of FinTech.

[Image Suggestion]: Insert a photo of your early handwritten notes or a screenshot of a messy Excel sheet from the "Data Finding Phase."

This month was about moving away from manual work and into automation and coding.

Week 1 (Nov 1–7): Overpass Turbo Power. I stopped manual Google searches and learned to use Overpass Turbo. I wrote a query to automatically pull every "bank," "atm," and "university" location in Malaysia directly into my system.
Week 2 (Nov 8–14): The Normalization Logic. I began writing my normalize(series) function in Python.
def normalize(series):
return (series - series.min()) / (series.max() - series.min())
This was crucial because you can't compare "RM 1 Million GDP" to "10,000 Students" without scaling them both from 0 to 1.
Week 3 (Nov 15–21): The Weighted Sum Model (WSM). I spent this week coding the Dynamic Weights. For the SME vertical, I gave Regional GDP a 50% weight ( $w=0.5$ ), while for E-Wallets, I shifted that weight to Mobile Usage.
Week 4 (Nov 22–30): The "Ocean" Logic. I coded the Strategic Cluster logic. I used percentiles to separate the districts: if potential is high and competition is low, the code labels it a "🔵 Blue Ocean".

This was the "fun" phase where the data finally started to look like a real app.

Week 1 (Dec 1–7): The "Twin District" Finder. I coded the Market Look-Alike module using Euclidean Distance.
dist = np.sqrt(((features_norm - target_vec) ** 2).sum(axis=1))
Now, if a user likes their results in Petaling, the system instantly finds similar districts like Johor Bahru.
Week 2 (Dec 8–14): Guarding the Budget. I built the Campaign ROI Simulator. I programmed a warning trigger: if a user's budget tries to target more than 5% of a district's population, a "High Saturation Risk" alert pops up.
Week 3 (Dec 15–21): The Persona Generator. I used my dataset to create a "Rule-Based Decision Tree." If a district has high income but high security usage, the app labels the audience "The Cautious Wealth Builder".
Week 4 (Dec 22–31): Streamlit Polish. I spent the final week of the year designing the Sidebar Control Panel and the Interactive Maps to make the app look professional.

Week 1 (Jan 1–7): Showtime. Jan 6th was Presentation Day!. I demonstrated the live DSS to the panel, showing them how the map shifts colors based on product focus.
Week 2 (Jan 8–14): The "Deep Dive" Revision. Based on panel feedback, I went back into my code to refine the Similarity Scores and double-check my MCDA justifications.
Week 3 (Jan 15–20): The Final Wrap-Up. I finished the final report. I made sure to emphasize that while my model is rule-based and transparent, future versions could use Real-Time API feeds to be even more accurate.

Nur Amyliana Syafika (e - Portfolio)