Automating Calibration and Report Generation with Google Colab: A Comparative Analysis of DHT11 and HTC-1 Sensors Using Fuzzy Logic and LaTeX

A Comparative Analysis of DHT11 and HTC-1 Sensors Using Fuzzy Logic and LaTeX

When I first started comparing the accuracy of my DHT11 sensor, I didn’t expect the project to evolve into a full data analysis pipeline. Initially, my goal was simple — determine how accurate the DHT11 really was compared to a reference instrument, the HTC-1. But as I went deeper, I realized I could automate the entire workflow: from analyzing raw data to generating a complete LaTeX-formatted PDF report, directly in Google Colab.

This post walks through how I designed that system — how I collected and analyzed the calibration data, applied fuzzy logic for comfort assessment, and produced a full scientific report automatically.

Introduction

Monitoring environmental conditions like temperature and humidity is critical for maintaining comfort and energy efficiency, especially in tropical office environments. Low-cost IoT sensors such as the DHT11 are convenient but not always reliable. My intention was to evaluate how well the DHT11 performs compared to a reference-grade device like the HTC-1 and to see if calibration could meaningfully improve its readings.

To make the process reproducible and less error-prone, I conducted the entire analysis in Google Colab, which offered a complete Python environment, built-in plotting, and direct Google Drive integration for storing results.

Data Collection

The experiment took place in a naturally ventilated office located in Indonesia. I placed both sensors — DHT11 and HTC-1 — side by side under identical conditions to minimize spatial variation. I collected two datasets: one before calibration and one after calibration, representing the performance difference.

The measurements covered several hours of normal office operation, capturing both day and evening temperature fluctuations. Each dataset contained paired readings of temperature and humidity.

Statistical Analysis in Colab

I started the analysis by importing necessary Python libraries and loading the datasets into Colab:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.metrics import mean_absolute_error, mean_squared_error

Error Metrics

The first evaluation used Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) to measure how far DHT11 readings deviated from the HTC-1 values. These two metrics provide a clear picture of how close the low-cost sensor’s readings are to the reference tool.

The equations I used were:

$$ MAE = \frac{1}{n} \sum |X_i - Y_i|, \quad RMSE = \sqrt{\frac{1}{n} \sum (X_i - Y_i)^2} $$

The MAE shows the average absolute error, while RMSE penalizes larger deviations more strongly.

Regression and Agreement

Next, I applied a linear regression model to characterize how DHT11 readings align with HTC-1 readings:

$$ Y = \beta_0 + \beta_1 X + \epsilon $$

This helped determine how calibration improved the sensor’s linear relationship. I also used Bland–Altman analysis to study the bias and consistency between the sensors. After calibration, the bias was notably smaller, and the readings were more uniform, confirming the improvement.

Calibration Outcome

After running the calibration procedure, the results were clear: the DHT11 sensor’s measurement accuracy improved dramatically for both temperature and humidity. While the pre-calibration readings showed noticeable offsets, the post-calibration values closely matched the reference measurements. This validated that even inexpensive sensors could achieve acceptable accuracy when proper calibration and statistical correction are applied.

Applying Fuzzy Mamdani Logic for Comfort Evaluation

To extend the analysis beyond numbers, I implemented a Fuzzy Mamdani inference system to interpret the environmental data in terms of thermal comfort.

The system categorized the environment as Cold, Comfortable, or Hot, based on fuzzy membership functions designed specifically for tropical conditions:

cold = fuzz.trapmf(temp, [20, 20, 24, 25.5])
comfortable = fuzz.trimf(temp, [25, 27.5, 30])
hot = fuzz.trapmf(temp, [29, 31, 34, 34])

The fuzzy logic approach allowed me to quantify comfort perception on a continuous scale from 0 to 10. The analysis revealed that most readings fell within the Comfortable range, suggesting that the room environment maintained stable and favorable conditions throughout most of the observation period.

This was a useful complement to the calibration analysis, providing a more intuitive interpretation of what the sensor readings meant in human terms.

Automating LaTeX Report Generation

After verifying the results, I wanted to present them in a professional and reproducible format. Instead of manually creating figures, equations, and descriptions in a separate document, I used Python’s pylatex library to automatically generate the full report.

Here’s an example snippet of how I structured the automation:

from pylatex import Document, Section, NoEscape

doc = Document("Scientific_Thermal_Comfort_Report")

with doc.create(Section("Results and Discussion")):
    doc.append("This section presents the calibration outcomes and fuzzy logic analysis performed in Colab.")

doc.generate_pdf(clean_tex=False)

This basic example was expanded to include all key sections: Introduction, Methodology, Results, Discussion, and Conclusion. The LaTeX code was dynamically constructed based on actual Colab results — including the computed metrics, equations, and descriptive text.

To ensure the file was easily shareable, I saved it directly to Google Drive:

from google.colab import drive
drive.mount('/content/drive')
doc.generate_pdf('/content/drive/MyDrive/Scientific_Thermal_Comfort_Report', clean_tex=False)

This made the report instantly accessible through a shareable Drive link once the notebook finished running.

Reflections

What began as a small calibration test turned into a complete research workflow combining data analysis, fuzzy logic modeling, and automated documentation. Running everything inside Google Colab helped me avoid setup issues while maintaining full reproducibility.

The integration of LaTeX generation was the most satisfying part — it transformed Colab from a simple coding notebook into a publishing tool. Every rerun could produce a new report, fully formatted and containing updated values.

This project reminded me how automation can enhance both accuracy and presentation in small-scale research. It also showed that with the right tools, even low-cost sensors like the DHT11 can deliver meaningful and interpretable insights when supported by data-driven calibration.

Source and Availability

The complete Colab-ready Python script is available here: GitHub Gist — DHT11 & HTC-1 Calibration Using Linear Regression and Fuzzy Mamdani

The automatically generated PDF report can be viewed on Google Drive: Scientific Thermal Comfort Report (PDF)