vjvlb;k'poi-iufysfdafz vugiulgb;ii;ug;jb;

lko;oihp;uhbp8hu;j.jh;ip;8iokjbkjbkjbjbjb,b

1. Overview of the overall work Name: Diganta Sikdar ID: 1906170

2. Overview of Dataset: The dataset has 227 e-nose data with 50 of them air, 97 of them fresh fish and 80 of them are formalin infused fish. Each data was taken for 3 minutes with 100 Hz sampling frequency. Data was collected from 6 MOX sensors using ADC of Arduino mega The dataset was trained on the 1 st dataset with 80-20 split for train and validation set.

3. Overview of Dataset: The 2 nd dataset was collected 1.5 months after the 1 st dataset. The time gap was for the sensors to get drifted. For this , sensors were kept in open space for that time. In this dataset, 86 were of fresh fish, 60 were of air and 75 were of formalin fish. The dataset was tested on the 2 nd dataset.

4. List of MOX sensors

5. Dataset description

6. The models were trained on batch 1 and then tested on batch 2. 8 features were extracted from each sensor. So, from 6 sensors there will be 48 features extracted. 2 of the features were steady state feature : maximal difference in resistance and the ration of maximal difference and minimum resistance value. The rest of the features were transient where the ema ( exponential moving average) was calculated using different alpha values[0.1,0.01,0.009] and the 2 features of minimal and maximum values were used.

7. Figure 1: a) Fresh fish b)Formalin Fish c)Air from the 1 st dataset Figure 2: a) Fresh fish b)Formalin Fish c)Air from the 2 nd dataset

8. Sensor Drift Sensor drift refers to the gradual deviation of a sensor's output from its true or calibrated value over time, even when the input conditions remain constant. This phenomenon can occur due to various factors, including: Aging of Components : Over time, the physical and electronic components of a sensor may degrade, leading to changes in its response characteristics. Environmental Influences : Factors such as temperature fluctuations, humidity, vibration, and exposure to chemicals can alter sensor behavior . Contamination : Accumulation of dirt, dust, or chemical residues on the sensing element can affect its performance.

9. Electrical or Mechanical Stress : Repeated use or exposure to high stresses can cause permanent changes in the sensor's structure. Calibration Loss : Some sensors require periodic calibration to maintain accuracy. Drift may occur if this is neglected or if the calibration reference itself is unstable.

10. Models used:

11. Results from the models Models Accuracy Precision Recall F1-score Air Formalin Fresh Air Formalin Fresh Air Formalin Fresh BiLSTM 53.85 100 44 67 32 91 37 48 59 48 ConvTran 50.68 45 63 56 81 50 71 Fusion network 64.71 74 54 68 48 51 88 59 52 77

12. Curves of the fusion model